Profil & Copyshop and Human-Machine Interaction718380/FULLTEXT01.pdf · and Human-Machine...

Robotics and Control Lab, no. 6 Doctoral Thesis

Semi-Automating Forestry MachinesMotion Planning, System Integration,and Human-Machine Interaction

SIMON WESTERBERG

UNIVERSITETSSERVICE

Profil & CopyshopÖppettider:Måndag - fredag 10-16

Tel. 786 52 00 alt 070-640 52 01Universumhuset

Department of Applied Physics and Electronics/Industrial Doctoral SchoolUmeå University, SwedenUmeå, 2014

Department of Applied Physics and ElectronicsUmeå University, SwedenSE-90187 UMEÅ

In collaboration with the Industrial Doctoral School, UmeåUniversity

Copyright c© Simon Westerberg, 2014.Except Paper I,c© IEEE, and Paper II,c© Wiley Periodicals.ISSN: 1654-5419ISBN: 978-91-7601-061-7E-version available at http://umu.diva-portal.orgPrinted by Print & Media, Umeå University, Umeå, 2014

Semi-Automating Forestry Machines:Motion Planning, System Integration, and Human-Machine Interaction

Simon Westerberg

AbstractThe process of forest harvesting is highly mechanized in most industrialized countries,with felling and processing of trees performed by technologically advanced forestry ma-chines. However, the maneuvering of the vehicles through the forest as well as the controlof the on-board hydraulic boom crane is currently performedthrough continuous manualoperation. This complicates the introduction of further incremental productivity improve-ments to the machines, as the operator becomes a bottleneck in the process. A suggestedsolution strategy is to enhance the production capacity by increasing the level of automa-tion. At the same time, the working environment for the operator can be improved bya reduced workload, provided that the human-machine interaction is adapted to the newautomated functionality.

The objectives of this thesis are 1) to describe and analyze the current logging processand to locate areas of improvements that can be implemented in current machines, and 2)to investigate future methods and concepts that possibly require changes in work methodsas well as in the machine design and technology. The thesis describes the developmentand integration of several algorithmic methods and the implementation of correspondingsoftware solutions, adapted to the forestry machine context.

Following data recording and analysis of the current work tasks of machine operators,trajectory planning and execution for a specific category offorwarder crane motions hasbeen identified as an important first step for short term automation. Using the methodof path-constrained trajectory planning, automated cranemotions were demonstrated topotentially provide a substantial improvement from motions performed by experiencedhuman operators. An extension of this method was developed to automate some selectedmotions even for existing sensorless machines. Evaluationsuggests that this method isfeasible for a reasonable deviation of initial conditions.

Another important aspect of partial automation is the human-machine interaction. Forthis specific application a simple and intuitive interaction method for accessing automatedcrane motions was suggested, based on head tracking of the operator. A preliminary in-teraction model derived from user experiments yielded promising results for forming thebasis of a target selection method, particularly when combined with some traded con-trol strategy. Further, a modular software platform was implemented, integrating severalimportant components into a framework for designing and testing future interaction con-cepts. Specifically, this system was used to investigate concepts of teleoperation and vir-tual environment feedback. Results from user tests show that visual information providedby a virtual environment can be advantageous compared to traditional video feedbackwith regards to both objective and subjective evaluation criteria.

Delautomatisering av skogsmaskiner:Rörelseplanering, systemintegration och människa-maskin-interaktion

Simon Westerberg

SammanfattningSkogsavverkningsprocessen är i hög grad mekaniserad i de flesta industriländer, där av-verkning och bearbetning av träd utförs med hjälp av högteknologiska skogsmaskiner. Förnuvarande sker dock framförandet av dessa fordon, likväl som manövrerandet av deras hy-drauliska kranar, genom kontinuerlig manuell styrning. Detta försvårar ett inkrementelltinförande av produktivitetsförbättrande åtgärder, eftersom föraren blir en flaskhals i pro-cessen. En föreslagen lösningsstrategi är att öka produktionskapaciteten genom att höjaautomationsnivån. På samma gång kan arbetsmiljön för operatören förbättras genom min-skad arbetsbelastning, förutsatt att interaktionen mellan människa och maskin anpassastill de nya automatiserade funktionerna.

Målen med denna avhandling är 1) att beskriva och analysera den nuvarande avverkn-ingsprocessen och att lokalisera de områden av förbättringar som kan genomföras i nu-varande maskiner, och 2) att undersöka framtida metoder ochkoncept som eventuelltkräver ändringar i såväl arbetssätt som i maskinens konstruktion och teknikanvändning.Avhandlingen beskriver utvecklingen och integreringen avflera algoritmiska metodersamt genomförandet av motsvarande mjukvarulösningar, anpassade till det sammanhangsom skogsmaskinerna befinner sig i.

Efter registrering och analys av maskinförarnas nuvarandearbetsuppgifter har planer-ingen och utförandet av en viss kategori skotarkranrörelser identifierats som ett viktigtförsta steg för automation på kortare sikt. Användning av metoden banbegränsad rörelse-planering (path-constrained trajectory planning) har visat att automatiserade kranrörelserpotentiellt kan ge en väsentlig förbättring jämfört med rörelser som utförs av erfarna män-skliga operatörer. En utvidgning av denna metod har utvecklats för att automatisera enutvald mängd rörelser även för existerande sensorlösa maskiner. En utvärdering av meto-den visar att denna metod är genomförbar vid avvikande initialtillstånd.

En annan viktig sida av automatiseringen är människa-maskin-interaktion. För attföraren enkelt ska kunna aktivera de automatiserade kranrörelserna presenteras en in-teraktionsmetod baserad på avläsning av förarens huvudrörelser. En preliminär inter-aktionsmodell härledd från användarexperiment visar lovande resultat för att kunna ut-göra grunden till en metod för att bestämma kranrörelsens slutposition. Vidare har enmodulär mjukvaruplattform implementerats, där flera viktiga komponenter integreras i ettramverk för utformning och testning av framtida interaktionskoncept. Detta system harsärskilt används för att undersöka begreppen fjärrstyrning (teleoperation) och återkop-pling med hjälp av virtuella miljöer. Resultat från användartester visar att visuell informa-tion som tillhandahålls av en virtuell miljö kan vara fördelaktig jämfört med traditionellvideoåterkoppling med avseende på både objektiva och subjektiva bedömningskriterier.

List of Publications

Parts of the contributions presented in this thesis have previously appeared in peer-reviewedjournals or conferences. Four of the publications, including one manuscript, are repro-duced in the second part of this thesis, and will be referred to by Roman numerals (I–IV).

I. U. Mettin, S. Westerberg, A. Shiriaev, and P. La Hera. Analysis of human-operatedmotions and trajectory replanning for kinematically redundant manipulators. InProceedings of the 2009 IEEE/RSJ International Conferenceon Intelligent Robotsand Systems, pp. 795–800, St. Louis, USA, Oct. 2009.

II. D. Ortíz Morales, S. Westerberg, P. La Hera, U. Mettin, L.Freidovich, A. Shiriaev.Increasing the Level of Automation in the Forestry Logging Process with CraneTrajectory Planning and Control.Journal of Field Robotics, Volume 31, Issue 3,pp. 343–363, 2014. DOI:10.1002/rob.21496

III. S. Westerberg, A. Shiriaev. Modeling of head orientation for applications in ma-nipulator control. 2014. Submitted manuscript.

IV. S. Westerberg, A. Shiriaev. Virtual environment-basedteleoperation of forestrymachines: Designing future interaction methods.Journal of Human-Robot Inter-action, Volume 2, Issue 3, pp. 84–110, 2013. DOI:10.5898/jhri.v2i3.145

iii

iv List of Publications

The following constitutes a complete list of papers published by the author, categorizedby subject:

Human-robot interaction in forestry machine applications

• S. Westerberg, A. Shiriaev.Modeling of head orientation for applications inmanipulator control . 2014. Manuscript.

• S. Westerberg, A. Shiriaev.Virtual environment-based teleoperation of forestrymachines: Designing future interaction methods. Journal of Human-RobotInteraction, 2:3:84–110, 2013. DOI:10.5898/jhri.v2i3.145

• S. Westerberg, I. Manchester, U. Mettin, P. La Hera, and A. Shiriaev. Virtualenvironment teleoperation of a hydraulic forestry crane. In Proc. 2008 IEEEInternational Conference on Robotics and Automation, pp. 4049–4054, Pasadena,USA, May 2008.

Trajectory analysis, planning and control for forestry machine cranes.

• D. Ortíz Morales, S. Westerberg, P. La Hera, L. Freidovich, A. Shiriaev. Path-constrained motion analysis. An algorithm to understand human performanceon hydraulic manipulators. Transactions on Human-Machines Systems, Submit-ted manuscript.

• D. Ortíz Morales, S. Westerberg, P. La Hera, U. Mettin, L. Freidovich, A. Shiri-aev. Increasing the Level of Automation in the Forestry Logging Process withCrane Trajectory Planning and Control . Journal of Field Robotics, Volume 31,Issue 3, pp. 343–363, 2013. DOI:10.1002/rob.21496.

• D. Ortíz Morales, S. Westerberg, P. La Hera, U. Mettin, L. Freidovich, A. Shiri-aev. Open-loop control experiments on driver assistance for crane forestrymachines. In Proc. 2011 IEEE International Conference on Robotics and Au-tomation, pp. 1797–1802, Shanghai, China, May 2011.DOI: 10.1109/ICRA.2011.5980266.

• D. Ortíz Morales, P. La Hera, U. Mettin, L. Freidovich, A. Shiriaev, S. Westerberg.Steps in trajectory planning and controller design for a hydraulically drivencrane with limited sensing. In Proc. 2010 IEEE/RSJ International Conference onIntelligent Robots and Systems (IROS), pp. 3836–3841, Taipei, Oct. 2010.DOI: 10.1109/IROS.2010.5649779.

• U. Mettin, S. Westerberg, A. Shiriaev, and P. La Hera.Analysis of human-operated motions and trajectory replanning for kinematically redundant ma-nipulators. In Proc. 2009 IEEE/RSJ International Conference on IntelligentRobots and Systems, pp. 795–800, St. Louis, USA, Oct. 2009.DOI: 10.1109/IROS.2009.5354362.

v

• U. Mettin, P. La Hera, D. Ortíz Morales, A. Shiriaev, L. Freidovich, and S. Wester-berg.Trajectory planning and time-independent motion control for a kinemat-ically redundant hydraulic manipulator . In Proc. 14th International Conferenceon Advanced Robotics, pp. 1–6, Munich, Germany, June 2009.

• P. La Hera, U. Mettin, S. Westerberg, and A. Shiriaev.Modeling and control ofhydraulic rotary actuators used in forestry cranes. In Proc. 2009 IEEE Interna-tional Conference on Robotics and Automation, pp. 1315–1320, Kobe, Japan, May2009. DOI:10.1109/ROBOT.2009.5152522.

• U. Mettin, P. La Hera, D. Ortíz Morales, A. Shiriaev, L. Freidovich, and S. West-erberg. Trajectory Planning for Kinematically Redundant Forestry Cranes.In Fourth Swedish Workshop on Autonomous Robotics (SWAR), Västerås, Sweden,Nov. 2009.

Obstacle avoidance and path planning among circular objects.

• S. Rönnbäck, S. Westerberg, K. Prorok.CSE+: Path planning amid circles. InProc. 4th International Conference on Robots and Agents, pp. 447–452, Welling-ton, New Zealand, Feb. 2009.

Environment reconstruction of logs from 3D data.

• Y. Park, A. Shiriaev, S. Westerberg, S. Lee.3D log recognition and pose esti-mation for robotic forestry machine, In Proc. 2011 IEEE International Confer-ence on Robotics and Automation, pp. 5323–5328, Shanghai, China, May 2011.DOI: 10.1109/ICRA.2011.5980451.

• Y. Park, A. Shiriaev, S.WesterbergRobust log recognition and pose estimationfor harvesting and logging of forestry machines, In Fourth Swedish Workshopon Autonomous Robotics (SWAR), Västerås, Sweden, Nov. 2009.

Trajectory planning and control of a simplified helicopter.

• S. Westerberg, U. Mettin, A. Shiriaev.Motion planning and control of an under-actuated 3DOF helicopter. In Proc. 2010 IEEE/RSJ International Conference onIntelligent Robots and Systems (IROS), pp. 3759–3764, Taipei, Oct. 2010.

• S. Westerberg, U. Mettin, A. Shiriaev, Y. Orlov.Motion planning and control ofa simplified helicopter model based on virtual holonomic constraints. In Proc.2009 14th International conference on advanced robotics, Munich, Germany, June2009.

vi List of Publications

Acknowledgments

This thesis marks the ending of a long and exciting journey. Now that land has finallybeen sighted, I will take the opportunity to express my gratitude to some of the peoplewho have made the trip possible and enjoyable:

to Anton, for leading the expedition and for launching the ship in which I have beentravelling,

to Uwe and Pedro, for teaching me how to navigate in open water,to Leonid, Daniel and all my other colleagues and friends, who have kept me company

and helped me stay the course through stormy weather,to the people in Komatsu Forest, IFOR, Skogstekniska Klustret and Sveaskog for their

support, for financing my expedition, and for keeping my shipseaworthy,to Petter, Benkt, and all my fellow travelers and friends at the industrial doctoral

school for letting me know I was not alone on the sea, for sharing your travel stories, andfor preparing me for going ashore on another continent,

and finally, to Kajsa, Vidar, and Loke, the stars on my night sky, who have accompa-nied me from afar and who will continue to shine on my path whenthis journey is overand a new one begins.

Simon WesterbergUmeå, Sweden, 2014

vii

viii Acknowledgments

Contents

Abstract i

Sammanfattning ii

List of Publications iii

Acknowledgments vii

Introduction 1

1 Background and Motivation 31.1 The logging process . . . . . . . . . . . . . . . . . . . . . . . . . . 41.2 Forestry machines . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.3 Operating a forwarder . . . . . . . . . . . . . . . . . . . . . . . . . 71.4 Incentives for automation . . . . . . . . . . . . . . . . . . . . . . . 7

2 Automation and Teleoperation 112.1 Teleoperation and supervisory control . . . . . . . . . . . . . .. . . 13

2.1.1 Classification of teleoperation systems . . . . . . . . . . .. 132.1.2 Cooperation modes in semi-automation . . . . . . . . . . . 14

2.2 Automation in forestry . . . . . . . . . . . . . . . . . . . . . . . . . 142.2.1 Current trends . . . . . . . . . . . . . . . . . . . . . . . . . 162.2.2 Future scenarios . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Analysis of Human-Operated Crane Work 213.1 Recording the data . . . . . . . . . . . . . . . . . . . . . . . . . . . 213.2 Distribution of work . . . . . . . . . . . . . . . . . . . . . . . . . . 213.3 Typical crane motions . . . . . . . . . . . . . . . . . . . . . . . . . . 223.4 Link usage statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

ix

x Contents

4 Motion Planning 274.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274.2 Forward and inverse kinematics . . . . . . . . . . . . . . . . . . . . 284.3 Limitations and constraints . . . . . . . . . . . . . . . . . . . . . . .294.4 Path-constrained trajectory planning . . . . . . . . . . . . . .. . . . 30

4.4.1 Non-redundant manipulators . . . . . . . . . . . . . . . . . 304.4.2 Redundant manipulators . . . . . . . . . . . . . . . . . . . 32

5 Human-Machine Interaction 355.1 Interacting with pre-planned trajectories . . . . . . . . . .. . . . . . 35

5.1.1 Trajectory selection methods . . . . . . . . . . . . . . . . . 355.1.2 Target point selection methods . . . . . . . . . . . . . . . . 35

5.2 Interaction methods for selecting target positions . . .. . . . . . . . 365.3 Selecting the optimal trajectory . . . . . . . . . . . . . . . . . . .. . 375.4 Minimizing the effects of input error . . . . . . . . . . . . . . . .. . 385.5 Human-machine collaboration . . . . . . . . . . . . . . . . . . . . . 39

6 Summary of Papers 43

Bibliography 46

Introduction

1 Background and Motivation

Based on a worldwide production of roundwood estimated at around 3.4 billionm3 eachyear (Ylitalo, 2012), the forestry industry contributes to1.2% of the global GDP. Theindustry has a global workforce of approximately 14 millionpeople1 (2006), more or lessequally divided between wood production, wood processing and paper/pulp industries(FAO, 2011; Renner et al., 2008).

The forestry sector plays a notably important role in the economy of many coun-tries with large afforested areas, particularly in northern Europe. Forest industry productsmake up a large share of the value of exported goods for countries like Finland (18% oftotal exports), Latvia (16%) and Sweden (11%) (Swedish Forest Agency, 2013; Ylitalo,2012). Some of the most dominant actors on the global forestry market are countries withlong traditions of forest management and utilization, suchas the Scandinavian countries,Canada, and Russia, who e.g. together contribute to more than half of the global exports ofsawn softwood (2011) (Ylitalo, 2012). In recent decades, the increased establishment ofindustrial plantation forests has introduced new actors from additional parts of the world,especially Asia and South America, and increased the competition on the global market(Carle et al., 2002). This trend is expected to continue as the demand for forestry productsincreases, both due to an overall population increase and increased demand for energy andmaterials from renewable resources.

As competition and productivity demands increase, technological development andmechanization in forest harvesting and processing are natural consequences. The amountof purely manual work has in many places been greatly reducedduring several stages ofautomation and in many industrialized countries, large parts of the processes from treesto end products are heavily, or even fully, automated.

This section aims to give an introduction to modern forest harvesting methods andthe current state of mechanization. It provides examples ofhow current technology isimplemented and used, as well as its limitations and incentives for future improvements.

1This number represents the formally registered workforce, with a large additional of seasonal, self-employed or otherwise unreported workers, which have been estimated to at least an additional 30-45 millionpeople (Renner et al., 2008).

3

4 1. Background and Motivation

1.1 The logging process

The forest is an important natural resource and wood is used as raw material for a diversityof products, ranging from board and paper to fabrics, chemicals and fuel. Regardless ofthe end product and manufacture process, the supply chain always begins in the forestwith the harvesting, orlogging, of the trees. The logging task can be divided into foursub-tasks.

Felling, where the tree is cut and felled to the ground. This task includes the decisionprocess of selecting which tree to cut and deciding the felling direction.

Delimbing, where the branches and top of the tree are removed from the stem. Tradi-tionally, only the stem has been utilized, but recently the slash (tops and branches)has attracted interest for use as bio-fuel, for instance in combined heat and powerplants (Demirbas and Balat, 2006; Viana et al., 2010).

Bucking, where the stem is cut into logs (also known as slashing or cross-cutting). Thelocation of each cut, i.e., the length of each log, is determined by a combinationof different parameters, such as tree length, stem diameter, and possible stem dam-ages or irregularities. To optimize profit, the cross-cutting decisions are also basedon information about current prices and the demand for different log dimensions(Murphy et al., 2004).

Off-road transportation, relocating the wood from the stump in the forest to a specificlocation by the road-side, from which the logs will eventually receive further trans-port, usually by trucks.

A few separate logging methods have emerged, differentiated by the order in which thetasks are performed and the location (at the stump or the roadside) where the delimbingand bucking tasks occur.

In whole-tree(WT) logging, all processing occurs after transport to the road-side.This is currently a common method in, e.g., North and South America. Incut-to-length(CTL) logging, or the shortwood system, on the other hand, both delimbing and buckingis performed at the stump, leaving processed logs for transportation. This is the dom-inating logging system in Europe, particularly in the Nordic countries. A third systemis tree-length(TL) logging, where the delimbing occurs after felling and the stems aretransported to the road-side for bucking and sorting. The difference between the systemsis summarized in Table1.1.

1.2 Forestry machines

In the 1950’s and 1960’s, the logging work in most developed countries was dominatedby the so-called motor-manual system, where the felling andlog processing stages whereperformed with chainsaws and manual tools, while horses or tractors were used for trans-port (MacDonald and Clow, 2006). The first commercially successful off-road vehiclesspecialized for logging tasks were machines for tree and logtransportation introduced

1.2. Forestry machines 5

SYSTEMTASKS PERFORMED

FELLING SITE ROAD-SIDE

Whole-tree (WT) logging Felling Delimbing,bucking

Tree-length (TL) logging Felling,delimbing

Bucking

Cut-to-length (CTL)logging Felling,delimbing,bucking

-

Table 1.1: Different types of logging systems.

in the late 1950’s and early 1960’s (Andersson, 2004). In Swedish forestry, the trans-portation task was almost completely mechanized at the end of the decade; during the1960’s, the use of horses for log transport decreased from 80% to less than 5%. Thesuccess of the specialized transportation vehicles paved the way for further development,and the construction of machines performing other logging tasks followed shortly. At firsteach machine specialized in performing a single task, such as the mechanical delimber.In the second half of the 1960’s, the first combination machines were developed, suchas the processor, which could take care of both the delimbingand bucking steps. Themajor mechanization stages took place in the 1970’s. A rapiddevelopment and sophisti-cation of earlier innovations ensued and by the end of the decade, forest machines had awidespread use in several countries, including Sweden, Finland and Canada. By 1990, thelogging process in these countries was more or less completely mechanized (Axelsson,1998; Halme and Vainio, 1998; Nordfjell et al., 2010).

Since the early 1990’s, no major change has occurred in the concepts and basicfunctionality of forestry machines. Currently, two different machine systems dominatethe market in the developed forest-intense countries2. For whole-tree logging, afeller-bunchercuts and fells several trees and places them in piles. Agrapple skidderdrags(skids) the full-length trees to the road-side where the processing takes place. The cut-to-length system consists of aharvesterand aforwarder (Figure 1). The harvester hasa knuckle-boom crane with a harvester head at the end, equipped with a chain-saw forcutting and bucking the tree, delimbing knives for removingbranches, and feed rollersfor grasping the tree prior to felling, as well as for feedingthe tree through the processingunit. The forwarder is normally a six- or eight-wheeled vehicle with articulated-steeringand a grapple-equipped hydraulic crane for loading the logsonto its cargo space (loadbunk).

The prevalence of the respective systems in a geographic region can partly be ascribedto tradition, but is also affected by forest properties, such as tree size distribution, type

2A number of additional machine systems exist, mainly specialized systems for e.g. steep or wet terrain. Afew such systems are described by Rummer (2002).


Figure 1.1: A Valmet 860 forwarder by Komatsu Forest.

of terrain etc. While the WT system may offer a slightly higher productivity in somesituations (Adebayo et al., 2007), the damages on the residual stand have been shownto be lower with the CTL system (Limbeck-Lilienau, 2003). More detailed discussionsabout the different methods and their respective advantages can be found in Akay et al.(2004); Pulkki (1997).

Many of the methods and results presented in this thesis are applicable to all differentkinds of forestry machines, as well as similar machines, particularly other heavy machin-ery manipulators, e.g., in construction, mining, or cargo transportation. However, dueto its exclusive use in Swedish forestry, the work in this thesis only considers the CTL(harvester/forwarder) system. Much of the recent researchefforts regarding technologicaldevelopment of CTL vehicles has been done on the harvester. The harvester operationconsists of a series of subtasks performed in rapid succession, sometimes performed inparallel with other tasks. According to Burman and Löfgren (2007), a harvester operatorperforms in average 24 functions and makes 12 decisions during the processing of a singletree (<1 min). The work performed by a forwarder operator is similar in that it consistsof a well-structured sequence of repetitive subtasks. However, the number of subtasksand decisions is smaller and the actions are less time-critical than for the harvester, whichmakes the forwarder an easier target for automation, in particular in a shorter term. Forthis reason, the forwarder has been used as a starting point when discussing and exempli-fying the methods and technologies used for both short-termand long-term automationthroughout the thesis.

1.3. Operating a forwarder 7

1.3 Operating a forwarder

While in the cabin, a forwarder operator performs tasks that belong to any of the followingthree main categories.

1. Decisions and planning. This is the mental work that is required in order to planand coordinate the other work tasks and to optimize their execution. On a higherlevel, the operator needs to plan the route through the harvested area, select theorder in which to pick up the logs, and plan for the number of assortments and theirlocation. This category also includes lower-level / short-term planning, like how toorient the timber during a crane motion in order to avoid obstacles.

2. Vehicle transportation. Moving the vehicle between the log piles along the trail, aswell as between the forest and the road-side.

3. Crane manipulation. Grasping, moving and releasing logs. This comprises a ma-jority of the operator’s active work in the machine.

A more detailed description of the sub-tasks involved in forwarding and a schematic de-scription of the work flow is presented in Figure 1.2.

Studies report that between 80% and 90% of the time spent in the cabin consists ofcrane manipulation for both harvesters and forwarders (Dvorák et al., 2008; Gellerstedt,2002). The cranes are hydraulic manipulators with the position of the end effector (grap-ple) determined by four variables (three angles of rotationand one linear displacementposition). Additionally, a rotator controls the horizontal orientation of the grapple. Tra-ditionally, the hydraulic cylinders associated with each of these variables are individuallycontrolled by some motion of one of two analog joysticks. Thejoysticks provide electri-cal signals that command the flow rate of the hydraulic control system, which is equippedwith an electro-hydraulic valve unit that controls the flow.These valves influence the mo-tion of the hydraulic actuators, and each link can be moved independently. The typicalmapping between the joystick and the forwarder crane is shown in Figure 1.3.

1.4 Incentives for automation

Although current forestry vehicles represent a high level of mechanization compared tojust a few decades ago, there still exist a few areas where different levels of automationcould be beneficial and changes in design, implementation orconcepts could improvedifferent aspects of the logging process.

As is commonly the case in industrialization and the adaption of technological in-novations, a major driving force behind the technological development in the forestrysector has been that of potential economic benefits (MacDonald and Clow, 1999). Whenit comes to continued development and partial automation ofthe work process, economicincentives are still important. The overall objective can be generally described as de-creasing the cost/output ratio. To this end, two different approaches can be taken. Firstly,one solution could be to increase theproductivity(output/hour). This will require an in-creased level of automation, since the production capacityof modern machines is large

81.

Background

andM

otivation

Load cycle Unload cycleTransport

Crane manipulation Planning and decisionsVehicle transportation

Figure

1.2:Sta

ted

iagra

mfo

rth

efo

rwa

rde

rw

ork

tasks.

1.4. Incentives for automation 9

Right joystickLeft joystick

In

Left Right

Out

In

Slewing

Telescope

Out

Outer

boom

Rotator

Left Right

Inner

boom

Up

DownClose

Open

Grapple

Figure 1.3: Typical dual joystick control of a forwarder crane.


enough to make the operators bottlenecks in the logging process (Hellström et al., 2009).Semi-automated functions can also reduce the productivitygap between beginners andexperienced drivers (Brander et al., 2004; Burman and Löfgren, 2007). Currently, it takesseveral years to become fully proficient and the productivity difference between a skilledand non-skilled operator can be as much as 25-40% (Löfgren, 2009). Secondly, econom-ical winnings can also be achieved by reducing theoperating costs(cost/hour). Also inthis case, automation can be beneficial. For instance, salaries currently makes up 30-40%of a forwarder’s running costs (Hellström et al., 2009), andwith more tasks possible toperform without human intervention, this cost can be reduced. Fuel cost is another largeportion of the total cost. Brunberg (2006) reports an average fuel consumption of11.8 l/hor, alternatively,0.7− 1.4 l for eachm3 of transported timber (Klvac and Skoupy, 2009).While human operators can learn some heuristic methods or guidelines to adopt an en-ergy efficient behavior, unlike a computer they lack the ability to mentally perform thecomplex real-time calculations that are required to find an optimal tradeoff between timeand energy efficiency in terms of minimized operating costs.

There is often a tradeoff between productivity and running costs. As the productiv-ity of individual machines in itself is no goal for the industry, it is possible to imaginea scenario where slightly less productive automated machines with reduced costs (e.g.salary and fuel) are cheaper and preferable alternatives tomore productive, but also moreexpensive, traditional fully manually controlled machines.

Theworking environmentis another relevant topic. The comfort of the operator hasbeen an important factor in recent machine development. Historically, the ergonomic sit-uation in both harvesters and forwarders has been far from ideal. The high workload inthe upper extremities during long hours of machine operation has been associated withdiscomfort and pain in both a short- and long-term perspective (Östensvik et al., 2008).In the 1980’s, the one year prevalence of neck or shoulder disorders among operators wasreported to be 50-80% (Attebrant et al., 1997). It has since improved, but still presents aproblem. In a study from 2005, the percentage of operators that often or very often expe-rienced pain in neck or shoulders was 16.2% and 15.5%, respectively (Vik, 2005). This isnot only a working environment issue, but also affects the productivity negatively, sincedrivers experiencing pain will not perform their best.Mental fatiguefrom the requiredconstant attention to the work tasks, enhanced by high levels of noiseandvibrationsap-parent in the cabin, results in lower performance during theend of a work shift (Synwoldtand Gellerstedt, 2003; Winkel et al., 1998). Further, an experienced driver that changesoccupation or retires earlier due to health problems may need to be replaced by an inex-perienced driver that will need several years to achieve a comparable productivity rate.

Additional issues that could be addressed by automation strategies are operator safety(e.g. risk of overturning vehicles) and environmental effects (e.g. fuel consumption andsoil damage).

2 Automation andTeleoperation

Parasuraman (2000) defines automation as “the execution by machine, usually a com-puter, of a function previously carried out by human.” Automation has for centuries beena way to increase productivity in production processes. In many traditional industrialsettings, the work process is often almost completely automated. Some examples arethe manufacturing and assembly processes of the automotive, machinery or electronicsindustries.

The incentives for automation can concern performance, where easily measurablequantities, such as precision, speed, or force, are improved. Robots and other kinds of au-tomation hardware can be stronger, more accurate, and more agile than humans. Robotsare more enduring and reliable, they do not suffer from performance decrease over thecourse of a work shift due to fatigue (Stentz et al., 2002). Automation is also introducedin order to improve the workers’ conditions. By replacing human labor in dangerous orexhaustive tasks, the human workforce can experience improvements in safety and work-ing environment and have a reduced workload. Typical application areas include mining(Larsson, 2011) or operation of underwater vehicles (Bingham et al., 2010). Machines orrobots do not disapprove of tedious or repetitive tasks. Automation of this kind can liber-ate human resources that are better suited for other tasks involving, e.g., decision making,programming or design.

Providing an exact definition of what automation includes isdifficult, but Sheridanand Parasuraman (2005) incorporates the following properties in the term:

1. Mechanized sensing of the environment

2. Automatic processing of data and decision making

3. Mechanical actions on the environment

4. Communication of processed information to the human

Automation rarely causes a complete removal of humans from the work process. Evenwhen all parts of a process are automated, a human operator usually has the role as a su-pervising agent. One of the reasons for this is to help resolving the issue of responsibility,in case an accident would occur.

11

12 2. Automation and Teleoperation

When introducing automation into a work flow where humans are involved some-where in the process, considerations need to be taken to the human perspective. It isimportant to understand how the human agents are affected bytheir new role in the workflow. Users of a partly automated system must understand the automated process wellenough to be able to interact with it in a correct, safe and efficient way.

Two partly overlapping research fields concerned with the interplay between humansand automated machinery arehuman-machine interaction(HMI) and human-computer in-teraction (HCI). While the two terms sometimes are used synonymously, HCI is more fo-cused on software-related interaction concepts by addressing “the design, evaluation, andimplementation of human-operated interactive computing systems” (O’Malley, 2007).HMI, on the other hand, assumes a more general type of machinefor which it regardsinteraction with both software and hardware.

Another more specific discipline ishuman-robot interaction(HRI), which is dedicatedto “understanding, designing, and evaluating robotic systems for use by or with humans”(Goodrich and Schultz, 2007). It differs from HCI and HMI in that it always includes aphysical dimension where a complex dynamic system (robot) is also interacting with thephysical (real-world) environment (Scholtz, 2002). This means that HRI to a further ex-tent needs to consider the processing (usually in real time)of information and knowledgeabout the environment and how to communicate this to the user. HRI also implies somelevel of autonomy and cognitive abilities from the system (Fong et al., 2003; Scholtz,2002). For a more general approach,human-automation interaction(HAI) can be definedas “the way a human is affected by, controls and receives information from automationwhile performing a task” (Sheridan and Parasuraman, 2005).

One important aspect of automation that is considered by these aforementioned disci-plines is that its introduction often affects the overall system in other ways than intended;careful considerations need to be taken also to possible downsides of the automation pro-cess. There exist a number of common issues encountered in automation. One of thechallenges is to create the right amount oftrust in the automated parts of the system.Lack of trust will make users hesitate to use automated functions, while over-reliance inthe system instead may result in lack of attention to the automated work task and the op-erator may even fail to prevent otherwise avoidable accidents. For in-depth discussionson these topics, see Parasuraman and Wickens (2008), Scholtz (2002), and Scholtz et al.(2004). Taking into account the respective strengths and weaknesses of humans and com-puters/machines/robots, it is apparent that different parts of the work flow benefit fromdifferent amounts of automation.

Attempts have been made to formalize this idea into a conceptcalledlevels of automa-tion (LOA), which specifies to what degree a robot can act on its ownaccord (Goodrichand Schultz, 2007). Different forms of LOA scales have been proposed, one of the ear-liest and most well-known is the one described by (Sheridan and Verplank, 1978), thatplaces an automated system on a scale of ten different degrees of automation, rangingfrom full manual control to full automation, along which tasks and responsibilities arebeing distributed in varying amounts between humans and machines. While this scalefocuses on decision support, similar scales for other task domains have been presented

2.1. Teleoperation and supervisory control 13

(Endsley, 1999; Kaber and Endsley, 2004; Trouvain et al., 2003).The LOA concept has later been extended into a discussion on another important as-

pect of automation, namely that different parts of a complexsystem can be automatedto different degrees. Parasuraman (Parasuraman, 2000) presents four important subtaskdomains that each can have its own level of automation:information acquisition, infor-mation analysis, decision and action selection, andaction implementation.

The choice of LOA is often a trade-off between different desired characteristics of thesystem. A low degree of automation promotes low initial cost, good situational aware-ness, high reliability, and easier recovery from automation failures, while a high degreeof automation can result in increased performance, improved working environment withlower workload, as well as a decreased running cost (Kaber and Endsley, 2004).

2.1 Teleoperation and supervisory control

Teleoperation means that a human operator is manipulating the physical world from adistant location (Kheddar et al., 2002). The performing agent of these manipulations hastraditionally been an articulated robotic arm; however, the concept is also applicable toremote control of mobile robots or other similar machines (Hannaford, 2000). Much ofthe early work in semi-automation originates in work on teleoperated robotic systems.Limited perception of the remote environment for the operator, as well as communicationdelays were common issues in these systems that could be addressed by an increased levelof autonomy.

2.1.1 Classification of teleoperation systems

Teleoperation systems for remote manipulation can be categorized according to the meth-ods by which they are controlled, i.e., the level of abstraction of the operator’s commandsto the system.

The first teleoperation systems were built around 1950, their task being to aid in nu-clear activities (Kheddar et al., 2002). Early systems consisted of two similar mechanicalarms: aslavearm located in the working environment, and amasterarm located at aremote location together with a human operator. The corresponding joints were mechan-ically connected, such that the slave arm exactly replicated the motion of the master arm(Figure 2.1(a)). By replacing the mechanical links with electronically controlled motorsand sensors, the control interface can be both physically and conceptually detached fromthe mechanical structure of the slave, and the human-machine interface can be reconfig-ured. Since the control still takes place in the joint space,the method is known asjointcontrol (Figure 2.1(b)). This corresponds to the conventional (non-teleoperation) controlmethod for many current manipulator-equipped heavy machine vehicles, such as excava-tors, construction cranes, and forestry machines.

In many cases, when controlling a robot arm, the specific joint angles during a mo-tion are not relevant for the execution of an assigned task, but only a means to realize themotions of the end effector. Hence, it is natural to considera telerobotic system where


the operator’s control input is directly correlated to the end effector configuration or mo-tion. This transfers the complexity of calculating the appropriate link configuration for aspecific end effector position, i.e. finding the solution to the inverse kinematics problem,to the computer controller. This control method is known asCartesian control(Figure2.1(c)). For cranes and other similar manipulators, this control method is sometimescalled crane-tip control or boom-tip control, or more generally end-effector control.

In the last category of control input methods, additional motion control functionalityis transferred to the computer. The continuous input signals are replaced with discretehigher-level commands that specify properties of the desired motion. This communica-tion can for instance consist of intermediate waypoints or final target positions for theend effector, which requires an automated planning and execution of the crane motion.In a scenario where the operator takes a more supervisory role, the communication mayconsist of some formalized description of some aims of the task, which requires a higherlevel of automation. This control concept (Figure 2.1(d)) is commonly denotedsupervi-sory control(Sheridan, 1995).

2.1.2 Cooperation modes in semi-automation

In a semi-automated system, some tasks are performed by humans, while some are per-formed by machines. This corresponds to the mid-range of theLOA scales discussedabove. In an optimal collaboration strategy between the human and the machine, eachagent is assigned tasks that they can proficiently perform, and the focus is on communi-cation and cooperation in order to optimize the overall performance (Kaber and Endsley,2004).

The cooperation in semi-automated systems is usually divided into two categories,depending on how the division of work is made between the human operator and themachine (Sheridan, 1992; Sheridan and Verplank, 1978). Thefirst is shared control,where the actors exercise simultaneous control or authority over (different parts of) thework process. The division of labor between operator and system could be performedbetween the DOFs of the robot, e.g., such that the operator controls the manipulator armon top of a mobile robotic platform that moves autonomously.The division could also bebetween tasks, e.g., the operator performs some task by controlling a robotic manipulatormanually, while an automated obstacle avoidance system ensures a collision free path.The second category istraded control, for which the division of labor is with respectto time, meaning that the actors work in turns. Here, no actions are being performedconcurrently except for during some transition period.

2.2 Automation in forestry

Compared to conventional robots, such as the industrial robots commonly utilized fore.g. manufacturing, processing, or packaging tasks, the development rate of automatedforestry robotic vehicles is much lower. The main limiting factor to the rate at whichautomation can be introduced in the field is the properties ofthe machines’ working en-vironment. While robots in assembly lines know more or less exactly what to expect for

2.2. Automation in forestry 15

Local Master Remote Slave

(a) Master-slave control

Local Control Device Remote Slave

(b) Joint control

Remote Manipulator DeviceLocal Control Device

x

y

x

y

(c) Cartesian control

Local Planning Remote Execution

(d) Supervisory control

(c) (d)

Figure 2.1: Control paradigms for teleoperation systems. (a) Specifying the motion of the slaverobot by replicating the configuration of a kinematically identical local master device. (b) Control-ling the individual link configurations by some human-machine interface. (c) Controlling the endeffector position in Cartesian space and having the computer resolve the necessary link positionsfor the desired position. (d) Communicating high-level commands to a semi-automated system.


each task, the forest environment is unstructured and cannot be known in advance. Thus,in order to achieve the same order of automation, a lot more sensing and information pro-cessing is required in order to safely and correctly performtasks that require interactionwith the surroundings. Further, the physical conditions are much more demanding than ine.g. a conventional factory setting. Physical equipment like sensors and electronics needto endure harsh weather condition with large variations in temperature and humidity. Al-gorithms for motion control and sensor data analysis need tobe adjusted according tovariations in external parameters, such as varied lightingconditions or changes in frictiondue to temperature shifts.

2.2.1 Current trends

The following constitutes a number of current trends in forestry machine technology andusage, as well as trends in the external environment and society, that are likely to affectthe priorities or directions of future forest machine development.

Sensors One important factor in more advanced automation is sensorsthat provide thesystem with feedback about its internal state, as well as sensors that allow the system toreact according to properties or changes in the environment. In the conventional manualcontrol system (Figure 1.3), it is the operator’s sensory information that allows continuousadjustments of each joint to achieve accurate crane motions. For an accurate computer-controlled low-level control of the crane linkage, sensorsthat can measure or estimate thejoint positions will be necessary.

The introduction of such sensors has become commonplace in the commercial pro-duction of, e.g., excavators or agriculture vehicles (Caterpillar, 2013). Lately, also forestrycrane producers have started to gain interest in including joint position sensors to open upfor more advanced functionality (Cranab AB, 2013).

In order for an autonomous system to be responsive to external stimuli, react to un-expected events, or plan and act according to external information, additional sensors arerequired. Currently, harvesters are equipped with sensor devices that measure e.g. treediameter and uses this information to assist the operator indeciding where to buck thetree for maximized profit. For an in-depth discussion on thistopic, see Marshall (2005).Contact-free optical versions of these sensors are currently considered (Andersson et al.,2008; Ärlestig, 2012). Though contact-free, they are stilllocal range sensors with thedetection space in a close proximity to the crane. A more complicated problem is long-range (several meters) detection, for example the analysisof stem dimensions while thetree still stands, which is desired for a more accurate log segmentation through correctselection of bucking points, or for creating local maps of the environment. A numberof attempts to estimate tree trunk diameters from a distancehave been made (Jutila et al.,2007; Ringdahl et al., 2013). This exemplifies a trend of increased interest in long-rangesensor devices for utilization in forestry. For reconstruction and detection of the nearbyenvironment, a rapid development of methods has taken place, both for forestry applica-tions and in other areas. Laser scanners, radars, and other remote sensing devices havebecome increasingly powerful, robust and affordable, and are popular for general naviga-


tion tasks, building local maps and detecting obstacles. For different applications in theharvesting and logging domains, A number of such applications have been suggested inthe forestry domain (Forsman and Halme, 2005; Kulovesi, 2009; Miettinen et al., 2007;Öhman et al., 2007; Park et al., 2011; Rossmann et al., 2009).

Crane-tip control The complexity and non-intuitiveness of the current joystick-basedcontrol method is a well-known problem. One solution that has been suggested for a longtime is crane-tip control, or Cartesian control (see Section 2.1.1). Having the operatorspecifying the 3D Cartesian coordinates of the crane tip instead of the 4D joint coordinateswould mean less joystick manipulation and a lower workload,as the coordination of thedifferent joints is automated. It also reduces learning time since control in task space ismore intuitive; a forward motion with the joystick corresponds to an extending (forward)motion of the crane tip. A number of control methods and working prototypes for crane tipcontrol have been demonstrated (Heinze, 2008; Löfgren, 2009; Parker et al., 1993; Prorok,2003) and simulation-based evaluations have been performed with regards to efficiency,workload, and learning rate (Brander et al., 2004; Egermark, 2005). Commercializationof this technique has been slow, however John Deere presented a harvester with crane-tipcontrol in early 2013 (John Deere, 2013a).

Redesigning machines With an increased demand for wood-based products follow in-centives to introduce mechanized harvesting in previouslyinaccessible forests. This in-cludes forests where only manual harvesting has been possible due to rugged or steepterrain, sometimes associated with great risks for the forestry workers.

To meet the new challenges of efficiently and safely traversing more inaccessibleterrain, new machine concepts and designs are considered, such as lighter and smallervehicles, or cabin-less and remote controlled vehicles (Milne et al., 2013). Differenttypes of locomotion are also under consideration to, e.g., achieve better stability in steepslopes. Walking machines, which have been considered for this purpose since the 1980’s(Halme and Vainio, 1998) but never reached production, havenow attracted new interest(SkogsElmia, 2013a). For wet and soft ground, the main problem is to reduce grounddamages, inspiring e.g. the new long-tracked bogie design (Edlund et al., 2013). Tak-ing into account the continuous rapid development of sensortechnology and automation,some concepts that have been suggested for future forest operations include machinesthat are equipped with several semi-automated boom cranes (Jundén et al., 2013), or for-warder trailers that can improve transport efficiency for longer distances (Lindroos andWästerlund, 2012).

2.2.2 Future scenarios

Many different scenarios can be presented where different levels of automation is usedto improve different aspects of harvesting and logging. Examples of automation featuresin the different functional dimensions of automation, using the definitions provided byParasuraman (2000) as presented above, are given in Table 2.2.2.


FUNCTIONAL

DIMENSION

LOW LOA H IGH LOA

Informationacquisition

Use sensors during operationto obtain raw data for off-lineanalysis

Replace raw data with filteredand processed information.

Ex. Recording crane motionsfor performance analysis.

Ex. Converting image data to3D objects in a virtualenvironment.

Informationanalysis

Interpolate and makepredictions

Integrate data, combine severalinput variables

Ex. During (manual) cranecontrol, assessing the risk forcollision with known obstacles

Ex. Planning a cranetrajectory that has desiredcharacteristics according to anoptimal tradeoff betweenexecution time, energy use,and mechanical wear.

Decisions andaction selection

Computer suggests a numberof alternatives that the humancan choose from.

The computer makes alldecisions but informs thehuman of its actions.

Ex: Presenting a number ofcrane trajectories that the userhas the option to select from.

Ex: Automatically selecting anappropriate crane trajectory.The user is informed about thedecision and given theauthority to stop the action.

Actionimplementation

Some activity from machineexecution, mostly manualcontrol

Mainly machine execution,with the exceptional manualintervention.

Ex: A few selected cranetrajectories can beautomatically executed.

Ex: Most crane manipulationtasks are automated, while theuser has a supervisory roleand controls the machinemanually in unexpected orunusual situations.

Table 2.1: Examples of automated functionality for different levels of automation.


While the list of theoretically feasible automation concepts is extensive, it is possibleto single out a few areas where potential benefits are large, by looking at methods or tech-nologies that either offer improvements to the current state and associated problem areas(see Section 1.4) or offer advantages in future forestry according to current trends (Sec-tion 2.2.1). Paper IV explores some of these future conceptsin more detail by describingthe design and potential use of a semi-automated forestry machine system for differentlevels of automation.

Below are some of the topics that are covered in the followingchapters of this thesis,as well as in the appended papers.

Teleoperation The difference between the current manual joint control andteleoper-ated joint control is, in many ways, only the physical distance between the operator andthe machine, since all the control input from the operator ismediated through a fullycomputerized control system. Short-distance (on-site) teleoperation could rather easilybe performed with current machines by separating the operator’s input device from themachine and adding a wireless communication signal betweenthe units. A prototype sys-tem,Besten, has been presented where a cabin-less harvester is remotely controlled froma forwarder (Bergkvist et al., 2006). While the potential performance of such a systemhas been challenged (Lindroos, 2009; Ringdahl et al., 2012), it demonstrates an importantaspect, namely that the relocation of the operator lessens some of the constraint on themachine design; current machine designs are largely user centered in order to create asafe and comfortable environment for the operator. A cabin-less forwarder has severalpotential benefits, as it can be more lightweight and have greater flexibility in size andgeometry.

Westerberg et al. (2008) presents a virtual environment system used to teleoperatea laboratory forwarder crane. In Paper IV, this system is used to evaluate a number ofdifferent interface methods for visual feedback during teleoperation.

Automated crane motions As mentioned in Section 1.4, the manual crane control is abottleneck when reaching for an increasingly efficient machine usage. Automated cranemotions would be a solution that also could improve the working conditions for the oper-ator.

Chapter 4 describes a method for planning trajectories thatcan be efficiently imple-mented and autonomously executed. The procedure is extended in Papers I and II. Forfurther discussions on this topic, see also Mettin et al. (2009a) and Morales et al. (2010).

Task-adapted human-machine interfaces Forestry machine producers have strong in-centives to continuously improve the user interfaces and interaction methods of the currentand future generations of machines, in order to improve the work situation for the opera-tors as well as the overall efficiency, and thereby create a selling point for the particularmachine producer. New or adapted methods will also be neededfor the changes in rolesand tasks that operators of future forestry machines will experience.

The suggestion of a new method for selecting target points for crane motions, intendedto be used in conjunction with automated motions, is presented in Paper III. This method


is based on tracking the head orientation of machine operators in order to estimate thedirection of their focus of attention. In addition, Paper IVinvestigates the use of virtualenvironments as means of presenting important informationto the users, both combinedwith different levels of automation and for providing high-quality visual or multi-modalfeedback during teleoperation.

3 Analysis of Human-OperatedCrane Work

One important part of the design of automated systems and human-machine / human-robotinteraction is to define and to analyze the current work process in order 1) to understandwhich domains and subtasks are best automated, 2) to identify potential issues that mayarise from automation, and 3) to appropriately integrate the automation into the organiza-tional structure. This section introduces such analysis for forwarder operation.

3.1 Recording the data

The analysis process starts with the recording of data. A forwarder (Valmet 830 by Ko-matsu Forest) was equipped with rapid prototyping hardware. A number of sensors wereattached to the hardware unit in order to record the following measurements:

• Crane joint position encoders forq1, q2, q3, q4,

• Hydraulic pressure, from which torque can be estimated.

• Control input from operator for all crane functionality (input corresponding to cranelink velocitiesq1, q2, q3, q4, rotator velocity, and grapple action) through the record-ing of analogue joystick signals,

• Head orientation of the operator, using an Xsens MTi inertial measurement unit(IMU), registering head orientation using integration of gyro signal (Xsens, 2014),

• External-view HD video recordings.

3.2 Distribution of work

From the recorded data, time intervals have been located that correspond to the individualsubtasks, or states, of the general work sequence presentedin the state diagram of Fig-ure 1.2. To speed up and structure the classification, an automatic processing procedurewas set up, based on the fact that properties of the sensor data are characteristic for each

21

22 3. Analysis of Human-Operated Crane Work

state. Experimentally found cut-off values were used both for individual data signals (e.g.slewing velocity, grapple input signal) or combinations ofmultiple signals (e.g. crane tipvelocity). These values were manually verified through animated 3D visualization of thedata, as well as playback of corresponding video recordings. The video recordings werealso used to analyze the states related to vehicle transportation. A sample segment of thecategorized data is shown in Figure 3.1. Since there is no wayto identify the temporallocation of the cognitive tasks, some states have been grouped together. Time distributionbetween the states is listed in Table 3.1. It should be noted that the distribution partlydepends on the properties of the harvested areas, such as distance to unloading site, treedensity and homogeneity (number of assortments), as well ason how log piles are orga-nized by the harvester operator (see Väätäinen et al. (2006)). However it should give arough estimate of the work distribution for general forwarder operation.

3.3 Typical crane motions

The workspace of the forwarder crane is large, and allows fora large diversity of cranemotions. In practice, however, the actual variety of motions performed during normaloperation is limited by, e.g.,

• the uniformity of the tasks with similar start and end positions for the end effector,

• specific “control patterns” that the drivers are taught during training,

• the properties of the motor learning process that consolidates similar actions forsimilar tasks,

• desired characteristics of the motions, such as speed, accuracy, and force,

• undesired characteristics that ought to be reduced or avoided, such as oscillations,collisions, or mechanical stress.

Examples of crane motions can be found in Figure 3.3, which shows the position of thecrane tip during a work session of 15 minutes from the recorded human-operated data.As can be seen, the majority of the end positions are confined to relatively small areas ateach side of the load bunk. Further, many of the geometrical paths to these points pathsare similar in appearance. This is advantageous in the case of an off-line motion planningprocess, which hence can be customized to produce a smaller number of highly optimizedmotions, as opposed to a large number of motions that covers the entire workspace.

3.4 Link usage statistics

The manual coordination of the different joints requires a high degree of joystick activ-ity. Figure 3.3 shows activity patterns of the joystick commands that correspond to theindividual joint variables. A large part of the complexity of the current input method isgenerated by the coordination between the different links.The number of joints moved

3.4. Link usage statistics 23

1,500 1,510 1,520 1,530 1,540 1,550 1,560 1,570 1,580

−2

0

2S2-S5

S6S7

S8S9

S6S7

S8S9

S2-S5S6

S7S8

time (t)

angl

e(r

ad)

q1q2q3q4

(a) Crane position data from joint encodersq1 − q4.

1,500 1,510 1,520 1,530 1,540 1,550 1,560 1,570 1,580

−1

0

1

2

·10−2

S2-S5 S6S7S8 S9S6 S7S8 S9 S2-S5S6 S7S8

time (t)

inpu

tsig

nal

(b) Joystick signal for rotator

1,500 1,510 1,520 1,530 1,540 1,550 1,560 1,570 1,580

−2

0

2

4

·10−2

S2-S5 S6S7S8 S9S6 S7S8 S9 S2-S5S6 S7S8

time (t)

inpu

tsig

nal

(c) Joystick signal for grapple

Figure 3.1: Recorded data signals, segmented by the different states from the state diagram (Fig-ure 1.2).


States n tseq(s) ttot(%)

Transportation 26.8

Load

cycl

e

S1 Move vehicle to loading area 7 162.4 15.0

S2-S5 Move vehicle to appropriate location 91 8.9 11.8

Crane manipulation 53.2

S6 Move grapple to logs 212 4.72 9.0

S7 Grasp logs 212 4.9 16.4

S8 Move logs to load bunk 212 7.5 15.7

S9 Release logs 212 4.9 12.1

Transportation 4.7

Unl

oad

cycl

e

S10 Move vehicle to unloading area 7 47.6 4.4

S11-14 Move vehicle to appropriate location 5 13.7 0.3

Crane manipulation 15.3

S15 Grasp logs 64 3.8 3.5

S16 Move logs to pile 64 7.1 4.4

S17 Release logs 64 4.3 3.8

S18 Move grapple to load bunk 64 4.9 3.7

Table 3.1: Time distribution between different states (subtasks) in the recorded forwarder work (seeFigure 1.2).n = number of occurrences of the state during the recorded time period;tseq= mediantime consumption for each sequence of the task;ttot= total time consumption for each state.

3.4. Link usage statistics 25

Figure 3.2: Typical motions of the end effector in the Cartesian 3D space. Recorded during 15minutes of regular forwarder work.

simultaneously is shown in Figure 3.4 for two of the recordedusers. It should be notedthat this result is based on the entire work cycle, includingthe transportation states, whichaccording to Table 3.1 make up approximately 31.5% of the total cabin time. During thecrane manipulation stages, the operators spend more than 55% of the time controllingthree or more of the six variables simultaneously.


q 2q 1

q 3q 4

q 5

10 20 30 40 50 60 70 80 90 100 110 120

time (s)

q 6

Figure 3.3: Visualization of joystick activity. Use of the different crane joints during two minutesof forwarding operation (blue=active, white=inactive), operator A.

0 1 2 3 4 5 60

10

20

30

40

Number of joints

Usa

ge(%

)

(a)

0 1 2 3 4 5 60

10

20

30

40

Number of joints

Usa

ge(%

)

(b)

Figure 3.4: Simultaneous manipulation of crane joints for two operators (A and B). The differentexperience of the operators (2 and 10 years) is reflected by the amount of simultaneous usage of thedifferent joints.

4 Motion Planning

According to the work distribution analysis in Section 3.2,the majority of the forwarderoperating time is spent on crane manipulation. It is therefore likely that any substantialgains in a semi-automated forwarder scenario require automated execution of a set ofcrane motions. This section gives an introduction to a motion planning method that canbe successfully applied to the forwarder crane to constructtime efficient motions thatcomply to the kinematic and dynamic constraints of the manipulator.

4.1 Definitions

In this thesis the following definitions are used:

Configuration A configuration of a manipulator is a complete specification of the locationof every point of the manipulator (Spong et al., 2006). The set of all possibleconfigurations is known as theconfiguration space.

Generalized coordinatesThe generalized coordinates, or configuration variables, are aset of independent variables that define the configuration ofa system. For a manip-ulator, such as the forwarder crane, the generalized coordinates consist of the jointvariablesqi, i = 1, 2, ..., n, wheren determines the number of degrees of freedom(DOF) of the system. The generalized coordinates of the forwarder crane are shownin Figure 1.3, Section 1.3.

Path A spatial description of the motion of a system. It can be described as a geometriccurve through either the configuration space or the operational (Cartesian) space.

Phase spaceThe phase space, or state space, of a dynamic system is the space of allpossible values attainable by the configuration variables and their derivatives. Apoint (q1, q1, q2, q2..., qn, qn) in the phase space represents a unique state of thesystem. The phase space of a one-DOF system is called aphase planeand hasdimension two.

Trajectory A trajectory is the time evolution of the state variables of adynamic systemalong a path and can be represented as a path (curve) through the phase spaceof the system. For a manipulator, each point along the trajectory is (directly or

27

28 4. Motion Planning

indirectly, depending on the representation) associated with positions, orientations,and velocities for each link.

4.2 Forward and inverse kinematics

The configuration of the crane (up to, but not including, the end effector) is represented bya four-dimensional vectorq = [q1, q2, q3, q4]T , see Figure 1.3. The end effector positionp = [px, py, pz]T described in the operational spaceR3 can be derived fromq using basictrigonometric functions, resulting in the forward kinematics function

p = ffk(q) (4.1)

The details and parameters of the crane kinematics specifiedusing the Denavit-Hartenbergconvention can be found in Paper I.

A manipulator is called kinematically redundant when the number of its DOFs islarger than what is necessary to execute a given task (Siciliano et al., 2009). For thetask of positioning the crane tip (the end of the telescopic boom), the forwarder crane isa kinematically redundant manipulator. The redundancy, illustrated by the difference indimensionality between the configuration space (S3 × R) and operational space (R3), isadded to provide the manipulator with a higher degree of dexterity and kinematic flexibil-ity.

The motion planning objective for a robotic manipulator is to appropriately place theend effector in a state, or series of states, such that it can be used to perform the intendedtask. The transformation from the operational space to the configuration space, i.e., froma desired target positionp⋆ to a target configurationq⋆ is known as the inverse kinematicsproblem, with an associated transformation function

q⋆ = fik(p⋆) (4.2)

For a redundant manipulator this is an underdetermined problem with an infinite numberof solutions. In the general case, individual solutions canbe found by iterative methods,such as cyclic coordinate descent (Kenwright, 2012; Wang and Chen, 1991), dampedleast squares methods (Wampler, 1986) or other local optimization methods, or methodsderived within the field of artificial intelligence, such as artificial neural networks (Kökeret al., 2004) or genetic algorithms (Nearchou, 1998).

If the positions or orientations of some of the individual links are relevant for thespecific task, this will impose restrictions on the set of desired solutions. In cases whereno or few such restrictions exist, such as for the forestry machine crane1, the additional

1More careful positioning of the individual links can be necessary or desired in some specific cases, mainlyto avoid obstacles, specifically obstacles of a certain height positioned along a direct line between the cranebase and the target position. However, due to the nature of the crane shape/structure (mainly positioned in orclose to a vertical plane) and the nature of the most common obstacles (tall and vertically standing trees), it isuncommon to encounter situations where one link configurationhits an obstacle whereas others do not.

4.3. Limitations and constraints 29

parameter can be used in the motion planning process to produce a target configuration(as well as configurations along the path) that gives some desired characteristics to themotion.

By taking the derivative of (4.1), we get

p = Jq (4.3)

whereJ = ∂ffk/∂q is the Jacobian matrix. One common approach is to use the (Moore-Penrose) pseudo-inverseJ+ as a basis for finding appropriate joint motions according tothe relation

q⋆ = J+p⋆ = (JTJ)−1JT p⋆ (4.4)

In its basic form, the pseudo-inverse method minimizes the norm of individual link veloc-ities. Extensions can be made to specify different secondary optimization criteria, suchas joint torque minimization (Hollerbach and Suh, 1987) or obstacle avoidance (Baillieul,1986). Westerberg et al. (2008) consider the pseudo-inverse method for the motion plan-ning and control of a forwarder crane, selecting the solution that minimizesqTP (q, c)−1q,where P is selected to be a positive definite weight matrix that combines a description ofthe desired contribution of the movement of each link with the reduction of velocities nearthe joint limits.

Such Jacobian-based solutions are local, i.e., they resultin locally optimized mo-tions with incremental movement from the current manipulator state (Beiner, 1997). Thismeans, for instance, that global characteristics for the motion reaching the target positionare not necessarily optimized. Further, such motions are not guaranteed to stay within thephysical limits of the machine, e.g., constraints on joint velocities may be violated.

4.3 Limitations and constraints

Every dynamic system has physical limitations that limit the values of its configurationvariables and their derivatives. For the forwarder crane, one example is the physicalrestrictions that introduce minimum and maximum limits to each joint. Another exampleis the limitations on joint velocities induced by the maximal oil flow rate of the hydraulicsystem. These limits have been experimentally identified byfollowing the procedureexplained in Paper II.

A system can also be limited, or constrained, in the relationbetween variables. Aconstraint is a general description of a limitation that canbe described as a (possiblytime-independent) function of the system state. If the constraint can be expressed as afunctionψ of the generalized coordinates (and possibly time), i.e.,

ψ(q, t) = 0, (4.5)

the constraint is calledholonomic(Choset et al., 2005). A classical example of holonomicconstraints is when two rigid bodies are physically connected in one point, i.e. coordinatesof the particular points are the same. The constraint does not limit the values of the state


variables, but defines a relation between them. Each linearly independent holonomicconstraint of the form (4.5) effectively reduces the dimensionality of the configurationspace by one, i.e., it “removes” one degree of freedom from the system.

Constraints that instead can be formulated in a form

ψ(q, q, t) = 0 (4.6)

that can not be integrated to (4.5) are said to benon-holonomic. Note that a constraint ex-pressed as a function of generalized velocities (derivatives of the generalized coordinates)is not necessarily non-holonomic. For instance, the constraint q1 = q2 can be integratedto the holonomic form asq1 − q2 + C = 0.

For hydraulically powered cranes velocity constraints arise naturally from the max-imum flow rate through the hydraulic system and are the main limiting factors in theimplementation of time efficient crane motions. Acceleration constraints, on the otherhand, are related to the maximum forces and torques that can be produced by the hy-draulic actuators. These constraints are configuration dependent and an identification ofthe constraints would require a system model with accurately identified parameters forboth friction and internal hydraulic dynamics. Such a modelis difficult to acquire, as forinstance temperature changes will produce large variations in friction parameters. Tunedcontrollers are instead used to validate the realizabilityof the desired input-output pair{q∗(t), u∗(t)}.

4.4 Path-constrained trajectory planning

The problem of manipulator trajectory planning often involves finding a motion that letsthe manipulator solve a specific task, or perform a motion between two given end points,in a way that is not only feasible but also optimal according to some performance criteria,which in general is a very complex problem.

One way to approach this problem is to solve each subtask separately, i.e., to firstfind a reasonable path and subsequently apply a procedure to assign configurations andvelocities along this path to create a trajectory with the desired performance. This isknown as thedecoupled approachor path-constrained trajectory planning(Bobrow et al.,1985b; Pfeiffer and Johanni, 1987; Shin and McKay, 1985).

To explain this method, let us first consider the non-redundant and non-singular case,where the generalized coordinates can be uniquely derived from the position of the boomtip.

4.4.1 Non-redundant manipulators

Assume that a path is given as a functionp(θ) = [x (θ) , y (θ) , z (θ)]T of a parametrizing

variableθ ∈ [θ0, θf ] that defines the position along the path2. For a non-redundant sys-tem, the inverse kinematics function (4.2) is uniquely defined from the inverse of (4.1),

2Any path that is provided in a different form representation, such as the unit circle in the(x, y) planedescribed in the implicit form,x2 + y2 = 1, should first be transformed to this format. In the case of the circle,the variable can be selected as the angle from the positivex-axis, such thatp(θ) = [cos (θ) , sin(θ), 0]T .

4.4. Path-constrained trajectory planning 31

so the configurationq at any position along the path can be described in terms of theparametrizing variable as

q = Φ(θ) = fik(p(θ)), (4.7)

provided the path is free from singular points. This path representation means that thefull state(q, q) of the system is now defined by the positionθ and velocityθ along thepath. A trajectory can be obtained by specifyingθ as a function of time, providing a timeevolution of the generalized coordinates asq(t) = Φ(θ(t)).

We will assume thatθ(t) is monotonic since the intervals between repeated valuesof θ can be removed without changing the path. Also, without lossof generality, let usassume thatθ > 0, i.e.θ(t) is monotonically increasing.

In order to select a feasible trajectory that conforms to thevelocity constraints of thesystem, individual joint velocity constraints can be projected to the(θ, θ) phase plane.The joint velocities can be derived from the parametrized joint function in (4.7) as

q = Φ′(θ)θ. (4.8)

Introducing the velocity constraintsqi,min and qi,max as the minimum and maximumvelocities for link i, the contribution from the individual link to the constraints on thevelocity along the path can be described by

θi,max = max

(

qi,max

φ′i(θ),qi,min

φ′i(θ)

)

, (4.9)

whereφi is thei’th element of the vector functionΦ. A feasible trajectory must at eachpoint conform to all of these constraints, i.e., the composite velocity constraint functionbecomes

θmax = mini(θi,max). (4.10)

The velocity constraints can be visualized in the phase plane, see Figure 4.1. A trajectoryis designed by shaping a functionf(θ) in the phase plane below the constraint function(4.10). The time evolutionθ(t) along the path is then found as the solution to the initialvalue problem

θ = f(θ), θ(0) = θ0. (4.11)

The execution timeT for the trajectory is given by

T =

θf∫

θ0

dθ

f(θ). (4.12)

The joint synchronization function (4.7) along the path, together with the time evolution ofthe parametrizing variable (4.11) gives the time evolutionof the generalized coordinatesas


0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

1

2

3

4

5

6

θ

dθ/d

t

q1q2q3

f(θ)

Figure 4.1: Velocity constraints in the phase space(θ, θ) for a 3DOF system. The dashed linesrepresent the positive constraints for each individual link. The shaded area is the region in whichone or more of the constraints is violated. A time efficient trajectory is designed by shaping afunctionθ = f(θ) close to the boundary of the shaded area.

q(t) = Φ(θ(t)). (4.13)

The choice off(θ) depends on the performance criteria. A common optimizationcriterion is the minimum execution time. In this case, the optimal result is achieved witha trajectory function that approaches the velocity constraints without violating them.

4.4.2 Redundant manipulators

For a manipulator with a degree of redundancyk, much of the procedure is similar asfor the non-redundant case. However, a number of additionalsteps are required to findappropriate parametrized joint configuration functions, corresponding to (4.7)n. The gen-eral procedure for defining such a function involves explicitly specifyingk of the variablesalong the path. For the 4-DOF forwarder crane,k = 1, which means that one of the jointsis defined as the redundant variable. Throughout this thesis, the evolution of the tele-scopic boom,q4, is selected and thus explicitly defined as a function of the parametrizingvariable, such that

q4 := φ4(θ). (4.14)

4.4. Path-constrained trajectory planning 33

The corresponding functions for the other three joints can now be derived from the kine-matics equations as

φ1(θ)φ2(θ)φ3(θ)

= ψ(p(θ), φ4(θ)). (4.15)

Consequently, the equation

q = Φ(θ) =

[

ψ (p (θ) , φ4 (θ))φ4(θ)

]

(4.16)

gives the complete parametrized configuration function in the form (4.7), after whichthe rest of the procedure (4.8)-(4.13) can be followed to give the trajectory as the timeevolution of the joint variables.

The choice of the functionφ4(θ) is an important step in the motion planning processfor the redundant manipulator, as it affects the set of possible trajectories. While thismethod always can generate the global time-optimal trajectory along a given path for anon-redundant system, the time efficiency for a redundant system is dependent on the se-lection ofφ4(θ). Preferably, this function is created as the result from an optimizationprocedure using the desired performance measure as the optimization criterion. As anexample, a time-efficient motion can be designed by representing the function as a Bézierpolynomial and search for the polynomial coefficients that maximize the permitted ve-locity θmax along the path, i.e., produces a trajectory with the minimalexecution timeaccording to (4.12).

5 Human-Machine Interaction

One advantage of computer-controlled crane motions is thatthe usage distribution amongthe links can be designed to, e.g., increase performance in terms of execution time orenergy efficiency, or to reduce mechanical wear. Once such motions have been designedand implemented, the next step is to construct an efficient interaction method that allowsthe operator to take advantage of the automated functionality.

5.1 Interacting with pre-planned trajectories

When utilizing an off-line motion planning procedure as described in Section 4.4, thesemi-automated system will have access to a library of pre-planned crane trajectories.Before the autonomous execution can take place, however, anappropriate trajectory mustbe selected. Unless the system by the means of sensing technology can make itself awareof the environment, some input is required that can convey the operator’s intentions. Eachinteraction method used for this purpose falls into one of the following two categories.

5.1.1 Trajectory selection methods

The first category consists of methods that present the user with a small fixed set of trajec-tories. From the available set, the operator will make an assessment of which trajectoryis most appropriate for the specific task and select it accordingly. Once the trajectoryis selected through some user interface, what remains of theautomated process is thestraight-forward execution of the crane motion, since all of the decision authority hasbeen retained by the operator. The challenges would insteadbe to decide the appropriatenumber and locations of the pre-defined target points in order to cover an essential part ofthe workspace while still being manageable for the operator, who is required to recall anddistinguish between the available trajectories.

5.1.2 Target point selection methods

Methods in the second category instead allow the user to define the desired target position(e.g. the position of the logs) freely within the workspace of the crane and let the systemselect the appropriate trajectory. The main advantage of such a method is that it has noinherent limitations in the number of trajectories that canbe made available and executed;

35

36 5. Human-Machine Interaction

the operator is only concerned with target positions, whilethe connection between thetarget point and the trajectory to be executed is performed by the computer. Thus, theresulting error between the end of the trajectory and the true target point can be reduced.This will help to reduce the amount of manual work required toreach the desired position,as well as potentially increase the performance. As an additional advantage, intelligentmotion selection strategies can be used to minimize errors since the system knows moreabout the intentions of the user. An example of such a strategy is described in Section 5.4.

5.2 Interaction methods for selecting target positions

An interaction method from the trajectory selection category requires each trajectory tobe readily available for the operator and to be assigned a unique, discrete input command.The corresponding user interface will provide this commandwhen triggered by a specificevent, such as the press of a button or a joystick motion. The number of such action-to-trajectory connections that can successfully be remembered and recalled by the user with-out additional effort is limited. Apart from tangible user interfaces, where the operatorinteracts with physical objects, the input can also be be mediated through other modali-ties, such as voice commands. While the environment is not optimal for this method dueto considerable levels of background noise in the cabin, a competent speech recognitionalgorithm could still be trained to recognize and distinguish between a small number ofcommands, as long as they are selected with this in mind and the commands are not toosimilar in the relevant characteristics used by the algorithm. Still, such methods may bebetter suited for other non-spatial and non-ordered commands used to describe separatefunctionality such as “grasp”, “open”, and “park”.

For target position selection methods, the conditions are different. The input data isno longer discrete or limited, but continuous and three-dimensional. While e.g. voicecommands can be constructed to select an item from a short list of commands, there isno intuitive way of using them to represent the 3D information required to specify anarbitrary target position. In this situation, some desiredcharacteristics of a user inputmethod for selecting target positions for a semi-automatedsystem would be

• intuitive, for easier learning and less mental workload,

• accurate, i.e., have a small average deviation from the intended target position tominimize unnecessary error correction and reduce the amount of manual work,

• adapted to the work in the actual environment that the machine operates in,

• providing access to the most relevant parts of the workspace,

• designed with considerations to ergonomic effects,

• comfortable for continuous use,

• requiring little effort to activate,

• fast, to not considerably delay the execution of the motion,

5.3. Selecting the optimal trajectory 37

• well integrated with manual control, i.e., the operator must be able to quickly re-sume manual control to perform the next subtask,

• subjectively preferred by the users compared to manual control.

One method that has the potential to satisfy many of these criteria is gaze tracking. It ishighly intuitive and requires no additional effort by the user, who in any case must lookat the target position before selecting it. A successful communication of this informationto the computer system would be a therefore be a useful, intuitive and efficient method ofselecting a target point. Paper III explores the possibility of estimating the gaze directionthrough head tracking, which is a simpler and cheaper solution than eye tracking. Initialuser tests show that a simple linear model of the head-to-gaze estimation function, care-fully calibrated according to the individual operator, canbe enough to give a reasonableand useful estimation of the gaze direction. Further, it presents an analysis of recordeddata from field tests that show that the results likely can be transferred to the forestrymachine setting.

5.3 Selecting the optimal trajectory

The library of pre-planned crane motions will provide access to a number of trajectoriesfrom a set of predefined crane configurationsC to a set of target pointsP . Assuming thata trajectory is specified by a time-dependent function

q(t) = [q1(t), q2(t), q3(t), q4(t)]T,

0 ≤ t ≤ T(5.1)

for the configuration variables(q1, q2, q3, q4), the motion library would then consist of aset of trajectories

Q = {q(t) | q(0) ∈ C, ffk (q (T )) ∈ P} , (5.2)

whereffk represents the forward kinematics of the crane.Any target position selection method will, as input to the computer system, provide

an estimationpest of the true (as intended by the operator) target positionptrue.Let P be the set of target points that are within the acceptable distanceL1 from the

estimated target positionpest according to some distance functiond1, i.e.

P = {p ∈ P | d1(p, pest) ≤ L1} . (5.3)

In the simplest case,d1 would correspond to the Cartesian distance between the points,d1(p, pest) = ||p− pest||.

Further, letC be the set of crane configurations that are within the acceptance limitL2 from the current configurationqc, according to some other distance functiond2, i.e.

C = {c ∈ C | d2(c, qc) ≤ L2} . (5.4)

One possible choice for the distance function would bed2(c, qc) = ||W (c− qc)||, whereW is a weight matrix that relates the deviation in initial conditions for each link to itscontribution to the overall accuracy of the trajectory.


From the set of acceptable trajectories

Q ={

q(t) ∈ Q | q(0) ∈ C, ffk(q(T )) ∈ P}

, (5.5)

the trajectoryq with the highest utility value according to some evaluationfunctionE willbe selected and executed,

q = argmaxq(t)∈Q

E(q(t)). (5.6)

Using a library of motions allows for complex time-consuming off-line optimization pro-cedures to find trajectories with high utility. The functionE can e.g. be related to com-pletion time, energy efficiency, or a combination thereof.

5.4 Minimizing the effects of input error

The selection method described by (5.6) will provide the optimal result for trajectorieswhereffk(q(T )) = ptrue, i.e. where the trajectory end points are identical to the intendedtarget position. In more realistic conditions, limitations on the number of motions in thelibrary as well as inaccuracies in the position estimation will introduce a discrepancybetween these values. In this case, an ideal utility function should instead evaluate theentire motion in its context, i.e.,

q = argmaxq(t)∈Q

E(q(t), qc, pest), (5.7)

whereqc is the current configuration of the crane. The variablesqc andpest are usedas arguments to the function since in order to evaluate the entire motion, including thetransition period from the initial configurationqc to a point on the planned path, as wellas from the end of the trajectory toptrue. The transition periods may include elements ofmanual control with properties that cannot be known in advance, but can be approximated.

When the target point is specified through a user interface, some error between theintended target pointptrue and the estimated target pointpest will occur. This generalproperty can be regarded as a result from the limited time allocated for communicatingthe target position to the system, a limitation that will make most user interfaces proneto errors, even if the magnitude of the error will differ. An additional part of the error isintroduced as a result from the discrepancy betweenffk(q(T )) ∈ P andpest. The sizeof this error will depend on the size of theP , as well as on the spatial distribution of itspoints.

For any specific end point, there is an infinite number of possible trajectories and itis possible, partly because of the redundant design of the manipulator, to find trajectoriesthat pass through different regions of the configuration space but still share the samedesired characteristics, e.g., have similar completion time. At a first glance, each of themmay seem an equally good choice. However, this assumes that the trajectories are fullyexecuted with no manual intervention, as can be the case ifptrue = pest. In reality, we canassume that the input specification method is imperfect and that generallyptrue 6= pest

5.5. Human-machine collaboration 39

holds. This inequality means that the operator likely will overtake the control task beforethe end point of the trajectory is reached, and (after a transition period of shared control)the operator will manually guide the end effector toptrue.

For any semi-autonomous system, it is important to rememberthat the overall per-formance of a task will be related to the added effects from the automated and manualparts. For a traded control system (see Section 2.1.2), the performance effects come froma combination of the automatic execution until the transition point; the transition period;and the manual control period. In order to select the correcttrajectory, it is necessaryto find the trajectory with the highest likelihood of having the best overall performanceaccording to this definition. In other words, even if a numberof trajectories are equallyefficient in terms of reachingpest, the same is not necessarily true for reachingptrue.

Figure 5.1 illustrates this concept by showing the impact onthe overall performancefrom selecting one of two possible trajectories to the same end pointpest. For each mo-tion, the required amount of manual work will be correlated to the distance from thetrajectory to the desired end pointptrue. Starting from the main objectives of simulta-neously increasing performance and improving the working conditions, it is desired tominimize the amount of manual control while maintaining high efficiency. Even thoughtraveled distance of the end effector does not translate directly to manual control time, itis apparent that a correlation exists, and that this distance is a relevant factor in the choiceof trajectory.

The “optimal” choice of trajectory thus depends on the true location of the desiredend pointptrue. This variable is unknown and can therefore not be used explicitly asa criterion/aspect of the trajectory selection process. However, it is possible to take aprobabilistic approach by using known statistical properties of the input error as well asan estimation of the cost associated with different parts ofthe error distribution.

5.5 Human-machine collaboration

As mentioned in the previous section, compatibility between HMI methods for manualcontrol and the execution of automated trajectories, as well as efficient transitions betweeninstances of automated and manual control, are important for the overall efficiency of asemi-automated system.

The specific details regarding the implementation of human-machine collaborationand traded control methods are in turn largely based on the implementation details of theexecution and control aspects of the automated motions, as well as on the specific HMImethods used for the respective tasks. The following properties, however, are important toconsider when designing and implementing a traded control method for this application:

• The transition must be soft. Abrupt changes in the velocity size or direction, eitherfor individual links or their effects on the end effector, may result in linkage oscil-lations and a hard-to-control swinging of the passive end effector. It may also havea negative influence on the durability of the machine.

• From the previous point follows that the transition must be gradual. The operator


m1

p1fE

(a) Trajectorym1, f = p1

m2

p1fE

(b) Trajectorym2, f = p1

m1

p2

fE

(c) Trajectorym1, f = p2

m2

p2

fE

(d) Trajectorym2, f = p2

Figure 5.1: Planned (solid red lines) and executed (dashed black lines) trajectories using tradedcontrol. Given the estimated target pointfE and two trajectoriesm1 andm2 ending at this point(taking different routes, but otherwise equivalent), the adequacy of each trajectory depends on thetrue fixation pointf . If f = p1, m1 gives the shortest path with least manual work, while iff = p2,m2 should be the preferred choice.

5.5. Human-machine collaboration 41

cannot be expected to resume a trajectory with the exact samelink velocities thatare used by the computer controller at the instance of transition.

• Following the transition period, the operator should be able to resume the normalcontrol method (i.e. joystick control) immediately, whichimplies that the joysticksare used during the transition period. There should be no “additional transitionperiod” between the transition period and the following period of manual control.

• The transition should be time efficient. Some time loss may beacceptable since itcan be compensated for by faster autonomous motions compared to manual control.But the productivity of the overall semi-autonomous systemshould not be less thanfor a conventional system.

• The transition period should be as short as possible in orderto increase the restingtime for the operator.

• The computer should not “fight” against the input from the operator in order to fol-low a planned trajectory. Instead, if the operator wants to deviate from the executedtrajectory, the system should comply with the proposed changes while graduallysubmitting the control authority to the operator.

• On the other hand, control efforts should not add up to something that breaks thephysical constraints in joint positions, velocities or accelerations.

• The transition should occur on the user’s initiative, and must be possible anywherealong the motion, not only at the end. The main reason for thisis safety, e.g., in casean obstacle must be avoided, but also to allow a possible change in the intentionsof the user.

6 Summary of Papers

Paper I: Analysis of human-operated motions and trajec-tory replanning for kinematically redundant manipulators

This paper describes the method of decoupled motion planning applied to the forwardercrane. In the first step of the planning algorithm, the geometric description of the desiredmotion is based on crane motions recorded during manual forwarding work by a humanoperator. New trajectories are replanned along the path by selecting new optimized jointprofiles, as well as by efficiently selecting a timing function that shapes the trajectoryclose to the velocity constraints of the crane.

Main contributions: Extraction and analysis of individual crane motions performedby human operators. Implementation of a motion planning algorithm that replicates themanual work optimized for time efficiency, as well as an optimization algorithm that si-multaneously searches for time-efficient paths and velocity profiles. Comparison betweenmanual and automated motions, demonstrating that the full potential of the machine is notused during manual crane work.

Paper II: Increasing the level of automation in the forestrylogging process with crane trajectory planning and control

This article outlines the construction of the hardware and software framework used forimplementing the motion planning and control methods for the forestry cranes, as wellas gives a short description of some control-theory specificaspects of the system. It ex-pands the trajectory planning method from Paper I to a sensor-less scenario for enablingshorter-term commercialization and introduction of the proposed algorithmic solution intocurrently available machines. A trajectory modification method is described where tra-jectories are adapted to provide a more consistent and accurate result when executed inopen-loop control mode.

Main contributions: The description of a trajectory modification method where tra-jectories are adapted to provide a more consistent and accurate result when executed in

43

44 6. Summary of Papers

open-loop control mode. Experimental tests that examine the performance in variousconditions, such as the invariability to load and reasonably deviating initial conditions.

Paper III: Modeling head orientation for applications inmanipulator control

The motion planning methods presented in Papers I and II provide the means for planningtime efficient crane trajectories that can be used to automatically control the crane to sometarget position. In a semi-automated scenario where the operator acts as the sensory inputfor external information (such as the structure and properties of the local environment),the operator needs to communicate his/her intentions to thesystem in order to initiatean appropriate trajectory for the given task. This paper suggests and describes a headtracking-based interaction method where an estimate of theoperator’s gaze direction isused to direct the crane to the desired target point.

Main contributions: The design and execution of a user experiment intended to eval-uate the feasibility of such an interaction method and to create a simple head-to-gazemodel that can be tested in a real-world context. Validationof the model by a comparisonto recorded head tracking data from regular forwarder operation.

Paper IV: Virtual environment-based teleoperation offorestry machines: Designing future interaction methods

This paper gives a detailed description of the system designand integration processes inthe development of a virtual environment-based interaction and teleoperation platform.Three different future scenarios are presented, representing different levels of automation(LOAs), see Figure 6.1. Each scenario provides its own requirements for the system de-sign, and the aim is to create a common platform to construct interaction methods thataid the operator across the range of LOAs. Further, the feasibility and possible advan-tages of VE-based human-machine interaction is examined both for on-site and off-siteteleoperation.

Main contributions: A formalization of the system design and integration using ascenario-based requirement analysis. Initial user-test experiments comparing differentfeedback methods using a number of both objective and subjective evaluation criteria.

45

Figure 6.1: Comparison of automation levels between the three presented scenarios.

46 6. Summary of Papers

Bibliography

Bibliography

Adebayo, A. B., Han, H.-S., and Johnson, L. (2007). Productivity and cost of cut-to-length and whole-tree harvesting in a mixed-conifer stand.Forest products journal,57(6):59–69.

Akay, A. E., Erdas, O., and Sessions, J. (2004). Determiningproductivity of mechanizedharvesting machines.Journal of Applied Sciences, 4(1):100–105.

Andersson, M., Hannrup, B., Larsson, W., Nyström, J., and Oja, J. (2008). Beröringsfrimätning kan ge bättre aptering (Noncontacting sensors for improved bucking). RE-SULTAT 16, Skogforsk, Uppsala. In Swedish.

Andersson, S. (2004). Skogsteknik förr och nu. InSkogshistoriska Sällskapets Årsskrift2004. Danagård Litho.

Ärlestig, M. (2012). A measuring device for a harvesting head. Munich, Germany:European Patent Office. European Patent No. EP 2,489,978.

Attebrant, M., Winkel, J., Mathiassen, S., and Kjellberg, A. (1997). Shoulder-arm mus-cle load and performance during control operation in forestry machines: Effects ofchanging to a new arm rest, lever and boom control system.Applied Ergonomics,28(2):85–97.

Axelsson, S.-A. (1998). The mechanization of logging operations in Sweden and its effecton occupational safety and health.Journal of Forest Engineering, 9(2):25–31.

Baillieul, J. (1986). Avoiding obstacles and resolving kinematic redundancy. InRoboticsand Automation. Proceedings. 1986 IEEE International Conference on, volume 3,pages 1698–1704. IEEE.

Beiner, L. (1997). Minimum-force redundancy control of hydraulic cranes.Mechatronics,7(6):537–547.

Bergkvist, I., Nordén, B., and Lundström, H. (2006). Bestenmed virkeskurir – ett inno-vativt och lovande drivningssystem (Innovative unmanned harvester system). RESUL-TAT 5, Skogforsk, Uppsala.

49

50 BIBLIOGRAPHY

Bingham, B., Foley, B., Singh, H., Camilli, R., Delaporta, K., Eustice, R., Mallios, A.,Mindell, D., Roman, C., and Sakellariou, D. (2010). Robotictools for deep waterarchaeology: Surveying an ancient shipwreck with an autonomous underwater vehicle.Journal of Field Robotics, 27(6):702–717.

Bobrow, J. E., Dubowsky, S., and Gibson, J. S. (1985). Time-optimal control ofrobotic manipulators along specified paths.International Journal of Robotics Research,4(3):3–17.

Brander, M., Eriksson, D., and Löfgren, B. (2004). Automation of knuckleboom workcan increase productivity. Results 4, Skogforsk, Uppsala.

Brunberg, T. (2006). Bränsleförbrukningen hos skördare och skotare vecka 13 och 39,2006. Arbetsrapport 629, Skogforsk.

Burman, L. and Löfgren, B. (2007). Human-machine interaction improvements of forestmachines. InForest Engineering Conference.

Carle, J., Vuorinen, P., and Del Lungo, A. (2002). Status andtrends in global forestplantation development.Forest Products Journal, 52(7/8):12–23.

Caterpillar (2013). New CatR© grade control system improves performance and produc-tivity of track-type tractors, motor graders, scrapers andexcavators. Press release.www.uk.cat.com/cda/files/2675502/7 (Accessed 2014-04-12).

Choset, H., Lynch, K. M., Hutchinson, S., Kantor, G., Burgard, W., Kavraki, L. E., andThrun, S. (2005).Principles of robot motion: Theory, algorithms, and implementa-tions. MIT press.

Cranab AB (2013). The first fully sensorized forwarder crane. Press Release.http://www.cranab.se/static/en/281/images/sensor_en.pdf (Accessed2014-04-12).

Demirbas, M. and Balat, M. (2006). Recent advances on the production and utiliza-tion trends of bio-fuels: A global perspective.Energy Conversion and Management,47(15):2371–2381.

Dvorák, Malkovský, J. Z., and Macku, J. (2008). Influence of human factor on the timeof work stages of harvesters and crane-equipped forwarders. Journal of Forest Science,54(1):24–30.

Edlund, J., Keramati, E., and Servin, M. (2013). A long-tracked bogie design for forestrymachines on soft and rough terrain.Journal of Terramechanics, 50(2):73–83.

Egermark, T. (2005). Boomtip control—a comparative evaluation of two control systemsfor forestry machines using a simulator. Master’s thesis, Linköping Universitet.

Endsley, M. R. (1999). Level of automation effects on performance, situation awarenessand workload in a dynamic control task.Ergonomics, 42(3):462–492.

BIBLIOGRAPHY 51

FAO (2011).FAO Yearbook of Forest Products. Number 46 in FAO Forestry Series. FAO.

Fong, T., Thorpe, C., and Baur, C. (2003). Collaboration, dialogue, human-robot interac-tion. In Robotics Research, pages 255–266. Springer.

Forsman, P. and Halme, A. (2005). 3-D mapping of natural environments with trees bymeans of mobile perception.IEEE Transactions on Robotics, 21(3):482–490.

Gellerstedt, S. (2002). Operation of the single-grip harvester: Motor-sensory and cogni-tive work. International Journal of Forest Engineering, 13(2):35–47.

Goodrich, M. A. and Schultz, A. C. (2007). Human-robot interaction: a survey.Founda-tions and Trends in Human-Computer Interaction, 1:203–275.

Halme, A. and Vainio, M. (1998). Forestry robotics – Why, whatand when. InAu-tonomous Robotic Systems, Lecture notes in control and information sciences, pages151–162. Springer.

Hannaford, B. (2000). Feeling is believing: History of telerobotics technology. InTherobot in the garden: telerobotics and telepistemology in the age of the Internet, pages246–274. MIT Press, Cambridge, MA, USA.

Heinze, A. (2008). Modelling, simulation and control of a hydraulic crane. Master’sthesis, Växjö University, School of Technology and Design.

Hellström, T., Lärkeryd, P., Nordfjell, T., and Ringdahl, O. (2009). Autonomous forestvehicles—Historic, envisioned and state of the art.International Journal of ForestEngineering, 20:33–38.

Hollerbach, J. M. and Suh, K. (1987). Redundancy resolutionof manipulators throughtorque optimization.Robotics and Automation, IEEE Journal of, 3(4):308–316.

John Deere (2013). John Deere’s intelligent boom control makes North American debut.Press Release.http://www.deere.com/wps/dcom/en_US/corporate/our_company/news_and_media/press_releases/2013/forestry/2013oct17_intelligent_boom_control.page (Accessed 2014-04-12).

Jundén, L., Bergström, D., Servin, M., and Bergsten, U. (2013). Simulation of boom-corridor thinning using a double-crane system and different levels of automation.In-ternational Journal of Forest Engineering, 24(1):16–23.

Jutila, J., Kannas, K., and Visala, A. (2007). Tree measurement in forest by 2d laserscanning. InComputational Intelligence in Robotics and Automation, 2007. CIRA2007. International Symposium on, pages 491–496. IEEE.

Kaber, D. and Endsley, M. (2004). The effects of level of automation and adaptive au-tomation on human performance, situation awareness and workload in a dynamic con-trol task.Theoretical Issues in Ergonomics Science, 5:113–153(41).

52 BIBLIOGRAPHY

Kenwright, B. (2012). Inverse kinematics–cyclic coordinate descent (CCD).Journal ofGraphics Tools, 16(4):177–217.

Kheddar, A., Chellali, R., and Coiffet, P. (2002). Virtual environment-assisted teleopera-tion. In Stanney, K. M., editor,Handbook of Virtual Environments: Design, Implemen-tation and Applications, chapter 48, pages 959–998. Lawrence Erlbaum Associates,Inc., Mahwah, NJ.

Klvac, R. and Skoupy, A. (2009). Characteristic fuel consumption and exhaust emissionsin fully mechanized logging operations.Journal of Forest Research, 14(6):328–334.

Köker, R., Öz, C., Çakar, T., and Ekiz, H. (2004). A study of neural network basedinverse kinematics solution for a three-joint robot.Robotics and Autonomous Systems,49(3):227–234.

Kulovesi, J. (2009). Motion vision based structure estimation in forestry environment.In Proceedings from the IEEE/RSJ International Conference onIntelligent Robots andSystems.

Larsson, J. (2011).Unmanned operation of load-haul-dump vehicles in mining environ-ments. PhD thesis, Örebro University, School of Science and Technology.

Limbeck-Lilienau, B. (2003). Residual stand damage causedby mechanized harvestingsystems. InProceedings of the Austro 2003 meeting: High Tech Forest Operations forMountainous Terrain, Schlaegl, Austria.

Lindroos, O. (2009). Bestensystemets konkurrensförmåga –Jämfört med vanligtskördare-skotaresystem. Technical report, Sveriges Lantbruksuniversitet (SLU).

Lindroos, O. and Wästerlund, I. (2012). Field study of a forwarder trailer concept –lower cost and fuel consumption at long distances. InProceedings of the Nordic Balticconference on forest operations – OSCAR 2012, volume 25, page 118.

Löfgren, B. (2009).Kinematic Control of Redundant Knuckle Booms with Automatic PathFollowing Functions. PhD thesis, Royal Institute of Technology (KTH), Stockholm,Sweden.

MacDonald, P. and Clow, M. (1999). ”Just one damn machine after another?” Techno-logical innovation and the industrialization of tree harvesting systems.Technology inSociety, 21(3):323–344.

MacDonald, P. and Clow, M. (2006). The industrialization oftree harvesting systems inthe Eastern Canadian forest, 1955–1995.Labour/Le Travail, 58:145–167.

Marshall, H. D. (2005).An investigation of factors affecting the optimal output log distri-bution from mechanical harvesting and processing systems. PhD thesis, Oregon StateUniversity.

BIBLIOGRAPHY 53

Mettin, U., La Hera, P. X., Morales, D. O., Shiriaev, A. S., Freidovich, L. B., and Wester-berg, S. (2009). Trajectory planning and time-independentmotion control for a kine-matically redundant hydraulic manipulator. InProceedings of the 2009 InternationalConference on Advanced Robotics.

Miettinen, M., Ohman, M., Visala, A., and Forsman, P. (2007). Simultaneous localiza-tion and mapping for forest harvesters. InProceedings from the IEEE InternationalConference on Robotics and Automation, pages 517–522.

Milne, B., Chen, X., Hann, C., and Parker, R. (2013). Robotisation of forestry harvestingin New Zealand–An overview. In10th IEEE International Conference on Control andAutomation (ICCA), pages 1609–1614. IEEE.

Morales, D. O., La Hera, P. X., Mettin, U., Freidovich, L. B.,Shiriaev, A. S., and West-erberg, S. (2010). Steps in trajectory planning and controller design for a hydraulicallydriven crane with limited sensing. InIEEE/RSJ International Conference on IntelligentRobots and Systems, pages 3836–3841.

Murphy, G., Marshall, H., and Bolding, M. C. (2004). Adaptive control of bucking on har-vesters to meet order book constraints.Forest Products Journal and Index, 54(12):114–121.

Nearchou, A. C. (1998). Solving the inverse kinematics problem of redundant robotsoperating in complex environments via a modified genetic algorithm. Mechanism andmachine theory, 33(3):273–292.

Nordfjell, T., Björheden, R., Thor, M., and Wästerlund, I. (2010). Changes in technicalperformance, mechanical availability and prices of machines used in forest operationsin Sweden from 1985 to 2010.Scandinavian Journal of Forest Research, 25(4):382–389.

Öhman, M., Miettinen, M., Kannas, K., Jutila, J., Visala, A., and Forsman, P. (2007). Treemeasurement and simultaneous localization and mapping system for forest harvesters.In 6th International Conference on Field and Service Robotics, volume 42, Chamonix,France. Springer.

O’Malley, M. K. (2007).Life science automation: Fundamentals and applications, chap-ter Principles of human-machine interfaces and interactions, pages 101–125. ArtechHouse Publishers.

Östensvik, T., Nilsen, P., and Veiersted, K. B. (2008). Muscle activity patterns in the neckand upper extremities among machine operators in differentforest vehicles.Interna-tional Journal of Forest Engineering, 19(2):11–20.

Parasuraman, R. (2000). Designing automation for human use: Empirical studies andquantitative models.Ergonomics, 43(7):931.

54 BIBLIOGRAPHY

Parasuraman, R. and Wickens, C. D. (2008). Humans: Still vital after all these yearsof automation.Human Factors: The Journal of the Human Factors and ErgonomicsSociety, 50(3):511–520(10).

Park, Y., Shiriaev, A. S., Westerberg, S., and Lee, S. (2011). 3D log recognition andpose estimation for robotic forestry machine. InProceedings from the 2011 IEEEInternational Conference on Robotics and Automation, pages 5323–5328.

Parker, N. R., Salcudean, S. E., and Lawrence, P. D. (1993). Application of force feedbackto heavy duty hydraulic machines. InProceedings of the 1993 IEEE InternationalConference on Robotics and Automation, pages 375–381. IEEE.

Pfeiffer, F. and Johanni, R. (1987). A concept for manipulator trajectory planning.IEEEJournal of Robotics and Automation, 3(2):115–123.

Prorok, K. (2003). Crane-tip control of a hydraulic crane: Anew approach. Technicalreport, Department of Applied Physics and Electronics, Umeå University.

Pulkki, R. (1997). Cut-to-length, tree-length or full treeharvesting.Central Woodlands,1(3):22–27.

Renner, M., Sweeney, S., and Kubit, J. (2008).Green jobs: towards decent work in asustainable, low-carbon world. UNEP.

Ringdahl, O., Hellström, T., and Lindroos, O. (2012). Potentials of possible machinesystems for directly loading logs in cut-to-length harvesting. Canadian Journal ofForest Research, 42(5):970–985.

Ringdahl, O., Hohnloser, P., Hellström, T., Holmgren, J., and Lindroos, O. (2013). En-hanced algorithms for estimating tree trunk diameter using2D laser scanner.RemoteSensing, 5(10):4839–4856.

Rossmann, J., Schluse, M., Schlette, C., Buecken, A., Krahwinkler, P., and Emde, M.(2009). Realization of a highly accurate mobile robot system for multi purpose pre-cision forestry applications. InProceedings of the 14th International Conference onAdvanced Robotics.

Rummer, B. (2002). Forest operations technology. General Technical Report SRS-53,Southern Forest Resource Assessment, USDA-Forest Service, Southern Research Sta-tion, Asheville, NC.

Scholtz, J. (2002). Evaluation methods for human-system performance of intelligent sys-tems. InMeasuring the Performance and Intelligence of Systems: Proceedings of the2002 PerMIS Workshop.

Scholtz, J., Antonishek, B., and Young, J. (2004). Evaluation of a human-robot interface:Development of a situational awareness methodology. InProc. 37th Annual Hawaii IntSystem Sciences Conf.

BIBLIOGRAPHY 55

Sheridan, T. B. (1992).Telerobotics, automation, and human supervisory control. MITPress, Cambridge, MA, USA.

Sheridan, T. B. (1995). Teleoperation, telerobotics and telepresence: A progress report.Control Engineering Practice, 3(2):205 – 214.

Sheridan, T. B. and Parasuraman, R. (2005). Human-automation interaction.Reviews ofHuman Factors and Ergonomics, 1(41):89–129.

Sheridan, T. B. and Verplank, W. L. (1978). Human and computer control for under-seateleoperators. Technical report, MIT Man-Machine SystemsLaboratory.

Shin, K. and McKay, N. (1985). Minimum-time control of robotic manipulators withgeometric path constraints.IEEE Transactions on Automatic Control, 30(6):531–541.

Siciliano, B., Sciavicco, L., Villani, L., and Oriolo, G. (2009).Robotics: Modeling, Plan-ning and Control. Advanced Textbooks in Control and Signal Processing. Springer-Verlag.

SkogsElmia (2013). Second attempt at a walking forest machine. http://www.elmia.se/en/wood/For-press/News-from-Elmia-Wood-rss/Second-attempt-at-a-walking-forest-machine/. (Accessed 2014-05-01).

Spong, M., Hutchinson, S., and Vidyasagar, M. (2006).Robot Modeling and Control.John Wiley and Sons, New Jersey.

Stentz, A., Dima, C., Wellington, C., Herman, H., and Stager, D. (2002). A system forsemi-autonomous tractor operations.Autonomous Robots, 13:87–104.

Swedish Forest Agency (2013).Swedish Statistical Yearbooks of Forestry 2013. SwedishForest Agency.

Synwoldt, U. and Gellerstedt, S. (2003). Ergonomic initiatives for machine operators bythe swedish logging industry.Applied ergonomics, 34(2):149–156.

Trouvain, B., Wolf, H. L., and Schneider, F. E. (2003). Impact of autonomy in multi-robot systems on teleoperation performance. InMulti-Robot Systems: From Swarmsto Intelligent Automata. Proceedings from the 2003 International Workshop on Multi-Robot Systems, volume 2.

Viana, H., Cohen, W. B., Lopes, D., and Aranha, J. (2010). Assessment of forest biomassfor use as energy. GIS-based analysis of geographical availability and locations ofwood-fired power plants in Portugal.Applied Energy, 87(8):2551–2560.

Vik, T. (2005). Working conditions for forest machine operators and contractors in sixEuropean countries. Number 25 in Rapport. Department of Forest products and mar-kets, Swedish University of Agricultural Sciences, Uppsala.

56 BIBLIOGRAPHY

Wampler, C. W. (1986). Manipulator inverse kinematic solutions based on vector formu-lations and damped least-squares methods.IEEE Transactions on Systems, Man andCybernetics, 16(1):93–101.

Wang, L.-C. and Chen, C. C. (1991). A combined optimization method for solvingthe inverse kinematics problems of mechanical manipulators. IEEE Transactions onRobotics and Automation, 7(4):489–499.

Westerberg, S., Manchester, I., Mettin, U., La Hera, P., andShiriaev, A. (2008). Vir-tual environment teleoperation of a hydraulic forestry crane. In Proceedings of the2008 IEEE International Conference on Robotics and Automation, pages 4049–4054,Pasadena, USA.

Winkel, J., Attebrant, M., and Wikström, B.-O. (1998).Konsensusrapporter rörande kun-skapsläget om arbetsmiljön i skogsmaskiner (Consensus reports concerning the stateof the art of the knowledge about the work environment in forest machines). Number 10in Arbete och hälsa. Arbetslivsinstitutet.

Xsens (2014). MTi.http://www.xsens.com/products/mti/. (Accessed 2014-05-01).

Ylitalo, E., editor (2012). Statistical Yearbook of Forestry. Finnish Forest ResearchInstitute.

Profil & Copyshop and Human-Machine Interaction718380/FULLTEXT01.pdf · and Human-Machine...

Documents

Transcript of Profil & Copyshop and Human-Machine Interaction718380/FULLTEXT01.pdf · and Human-Machine...