Graphical State-Space Programmability as a Natural Interface for Robotic Control

Graphical State-Space Programmability as a Natural Interface for Robotic ControlAnqi Xu, Gabriel Charette, Junaed Sattar, Gregory Dudek

McGill University, Canada

Thursday, May 6th, 2010

OverviewExisting robot programming techniques

Graphical State-Space Programming (GSSP)

Implementation on a representative robotic platform

User StudyAnqi Xu, McGill UniversityAqua: A Family of Underwater RobotsSurvey health of marineecosystems

Location-specific tasks:Take picture of this coralSearch for that type of fish

Infrastructure representative of most robotic platforms

Anqi Xu, McGill UniversityRobot Programming InterfacesTextual programming

Anqi Xu, McGill UniversityRobot Programming Interfaces (cont.)Textual programmingData flow programming (e.g. visual front-end for code)Lego Mindstorms NXT-G Software

Anqi Xu, McGill UniversityRobot Programming Interfaces (cont.)Textual programmingData flow programming (e.g. visual front-end for code)Graphical control / tele-operationProcerus Virtual Cockpit Software

Anqi Xu, McGill UniversityTask SpecificationTextual & data flow programming: if-then-else / switchFull control over sequence of executionAbility to specify branching pathsNested clauses hard to debug

Graphical control: on-screen command selectionCommands placed at distinct locations on mapNo path branching capabilitiesLimited selection of tasksAnqi Xu, McGill UniversityInsert waypoints through GUI

Location-specific tasks: standard computer code attached to waypointsBranching & looping pathsNatural visualization of data flow

Core robot AI code decoupled from path itinerariesMinimal overlap of functionalitiesHigh-level robot interactionGraphical State-Space Programming (GSSP)Anqi Xu, McGill University

GSSP Implementation onboard AquaAnqi Xu, McGill UniversityExisting development processLong compilation timesMonolithic codebaseLimited on-boardprocessing capabilities

Revamped development processScripts interpreted at runtime using Python API wrapperModular client-server designUpgradable client-side processing capabilities

Location-specific Python scripts

Load new paths within scriptsalternating routescontingency plansetc.

Create / modify paths at runtimebased on current state & latest observationsPath Generator GUIAnqi Xu, McGill University

User StudyGraphical state-space programming versus textual scripting

Trajectory through all marked locationsEach icon corresponds to a specific action commandFor textual scripting, use additional function calls to move robot5 subjects, 4 sessionsAlternate between GSSP & textual interfaces

Evaluation metric: elapsed duration

Anqi Xu, McGill UniversityStudy ResultsGSSP interface consistently faster than procedural coding

GSSP has minimal learning overhead

GSSPs high-level interaction prevents path-related & other simple programming errors

Anqi Xu, McGill UniversityConclusionGraphical state-space programming paradigmGeographical visualization of location-specific commandsCode for core AI routines decoupled from path itineraries

Advantages:Smaller codebaseFaster development timeHigh-level robot interaction

Ongoing work:Provide simultaneous visualization of branching pathsIntegrate with other robotic platforms

Anqi Xu, McGill University