Planning Under Uncertainty for Unmanned Aerial Vehicles...
Transcript of Planning Under Uncertainty for Unmanned Aerial Vehicles...
![Page 1: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/1.jpg)
Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)
by Ryan Skeele
Advised by Geoff Hollinger
Thesis Defense for the Masters of Science in Robotics
1
![Page 2: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/2.jpg)
UAVsSuccessfullyUsedforManyTasks
2
![Page 3: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/3.jpg)
3
UAVsOperateinChallengingEnvironments
3
![Page 4: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/4.jpg)
TheChallenges
4
![Page 5: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/5.jpg)
TheChallenges
● Precision sensing ability, and/or robust planning algorithms
● Coordination between UAVs, other
robots, and humans become increasingly important
● Planning in unknown or dynamic areas
5
![Page 6: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/6.jpg)
ThreeMainResearchAreas
Sensing Coordinating Planning
6
![Page 7: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/7.jpg)
Outline
Sensing Coordinating Planning
7
![Page 8: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/8.jpg)
RelatedWork
Sensing Coordinating Planning
8
● “An Unmanned Aircraft System for Automatic Forest Fire Monitoring and Measurement,” Merino, et al. 2012.
● “Planning periodic persistent monitoring trajectories for sensing robots in
gaussian random fields,” Lan & Schwager, 2013.
![Page 9: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/9.jpg)
Sensing
● Sensing stationary targets is well studied
● Dynamic targets are tricky
● Used for environmental monitoring
● One of the most challenging monitoring problems is wildfires
● The uncertainty of where the fire will spread is challenging
9
![Page 10: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/10.jpg)
10
Domain
Wildfires
● Dangerous for human pilots to get close
● Aerial sensing provides critical information
● Highly dynamic points of interest
10
*Taken from weathernetwork.com
![Page 11: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/11.jpg)
ProblemFormulation
● Fireline intensity is crucial information
● Regions of high intensity are dangerous
● Identify regions as hotspots
● Minimize the max time that a hotspot is left unattended
11
Where ! is max time untracked.
![Page 12: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/12.jpg)
SimulationEnvironment
● FARSITE generates fire characteristics
● UAVs limited to flying around fire
● UAVs have sensing radius
12
![Page 13: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/13.jpg)
SimulationSetup
K-means clustering is run on a frontier filtered for the highest intensities. K is determined by
13
where N is number of points of interest.
![Page 14: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/14.jpg)
Algorithm:Baseline
● Periodic monitoring
● Minimizes time untracked of all points
along the frontier
● Assumes no knowledge of fire
14
![Page 15: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/15.jpg)
Algorithm:ClusteredWeighted-Greedy
● Hotspot priority determined by time left untracked and distance from the UAV.
● A weighting parameter (") is applied to the travel cost (C) of each hotspot (#) to combine with the time ($ ) metric.
15
Weighted-Greedy
hotspots
![Page 16: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/16.jpg)
SimulationResults
● Compare Baseline and Weighted-Greedy
● Three different hotspot thresholds used, (.25, .35, .45).
● Seven different fires
16
Where ! is max time untracked.
![Page 17: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/17.jpg)
SimulationResults
17
Low
er is
better
J(t
) J(t) = sum of the max time untracked of all hotspots
Hotspot threshold = .25
![Page 18: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/18.jpg)
SimulationResults
18
J(t
)
Low
er is
better
J(t) = sum of the max time untracked of all hotspots
Hotspot threshold = .35
![Page 19: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/19.jpg)
SimulationResults
19
Low
er is
better
J(t
) J(t) = sum of the max time untracked of all hotspots
Hotspot threshold = .45
![Page 20: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/20.jpg)
HardwareImplementation
● FARSITE simulated fire
● Ground Station fed “live satellite data” from simulation
● Tethered IRIS+ Quadcopter
● 10 minute experiment
20
![Page 21: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/21.jpg)
HardwareImplementation
GPS trajectory with fire 2
1
a b
c d
![Page 22: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/22.jpg)
Discussion
● Naive methods of monitoring the fire miss valuable information
● More sophisticated sensing provides better results
● Fewer hotspots do a worse job representing the frontier regions
● Possible to bring this technology to the real world ● Publication:
○ R. Skeele, and G. Hollinger. "Aerial Vehicle Path Planning for Monitoring Wildfire Frontiers." Field and Service Robotics, 2015.
22
![Page 23: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/23.jpg)
Outline
Sensing Coordinating Planning
23
![Page 24: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/24.jpg)
RelatedWork
Sensing Coordinating Planning
24
● “Multi-Robot Coordination with Periodic Connectivity,” Hollinger & Singh, 2010.
● “The Sensor-based Random Graph Method for Cooperative Robot
Exploration,” Franchi et al., 2009.
![Page 25: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/25.jpg)
Coordination
● Is task allocation among multiple systems
● Provide robustness and fault tolerance
● Operate effectively in groups or with humans
● Better efficiency
● Requires communication protocols
● Coordination is difficult if the space is unknown
25
![Page 26: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/26.jpg)
Domain:IndoorExploration
High-impact applications:
● Urban search and rescue
● Industrial inspection
● Military reconnaissance
● Underground mine rescue operations
26
Key Challenges:
● Communication is uncertain
● Real robots have limited battery
● Planning through unknown maps
*Image from movie Indiana Jones
![Page 27: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/27.jpg)
ProblemFormulation
● Indoor environment represented in ℝ2
● Multiple UAVs merge maps
● Maximize the map returned
map to base station
27
MaximizeAreaMapped:
?
Area explored (Ar) with paths (%) in the possible path space (&), such that the path of each UAV (k) is less than the battery limit (B).
![Page 28: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/28.jpg)
SimulationSetup
● UAVs are modeled as discs with omnidirectional sensors
● Kinodynamics of the vehicles are not considered
● Each UAV has a limited battery life
● Communication is constrained by distance and obstacles
28
Simple Tunnel Map (50m x 50m)
Complex Office Map (120m x 40m)
![Page 29: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/29.jpg)
Algorithm:Baseline(FrontierExploration)
● Finds open cells next to unknown cells
● Uses blob detection to identify frontier regions
● Assigns robot to explore nearest frontier region
29
*Taken from robotfrontier.com
![Page 30: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/30.jpg)
● Coordinates robots to share information
● Robust to unreliable communication
● Considers limited battery life
Algorithm:EMSR(ExploreMeetSacriOiceRelay)
30
![Page 31: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/31.jpg)
● Coordinates robots to share information
● Robust to unreliable communication
● Considers limited battery life
Algorithm:EMSR(ExploreMeetSacriOiceRelay)
31
![Page 32: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/32.jpg)
● Coordinates robots to share information
● Robust to unreliable communication
● Considers limited battery life
Algorithm:EMSR(ExploreMeetSacriOiceRelay)
32
![Page 33: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/33.jpg)
● Coordinates robots to share information
● Robust to unreliable communication
● Considers limited battery life
Algorithm:EMSR(ExploreMeetSacriOiceRelay)
33
![Page 34: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/34.jpg)
● Coordinates robots to share information
● Robust to unreliable communication
● Considers limited battery life
Algorithm:EMSR(ExploreMeetSacriOiceRelay)
34
![Page 35: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/35.jpg)
SimulationResults
● 200 Simulations
● Random Starting Point
● Speed 1 m/s
35
![Page 36: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/36.jpg)
SimulationResults2 UAVs
36
Complex Office Map (120m x 40m)
Pro
porti
on o
f Map
Exp
lore
d
![Page 37: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/37.jpg)
4 UAVs
SimulationResults
37
Pro
porti
on o
f Map
Exp
lore
d
Complex Office Map (120m x 40m)
![Page 38: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/38.jpg)
8 UAVs in a simple tunnel
SimulationResults
38
Pro
porti
on o
f Map
Exp
lore
d
Simple Tunnel Map (50m x 50m)
![Page 39: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/39.jpg)
SimulationResults● Can use other exploration
techniques with our state machine on top
● Improvements range from 5% to 18%
● Better results with a larger team
Average of 200 simulation runs, with random start points.
39
![Page 40: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/40.jpg)
HardwareTesting
● Low cost platform, less than $1500
● Small enough to fit through doors
● Onboard vision and planning real time
● ROS planning and SLAM packages
● PX4 flight controller
● Xtion RGB-D Sensor
40
![Page 41: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/41.jpg)
HardwareTesting
Demonstrated with two quadcopters
● Standard ROS-Packages for navigation
● Complete onboard autonomy
● Un-instrumented environment
41
![Page 42: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/42.jpg)
HardwareResults
42
![Page 43: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/43.jpg)
Discussion
● First looked at single vehicle constrained planning
● Coordinated exploration outperforms non coordinated methods
● Indoor exploration is feasible
● Publication: ○ K. Cesare, R. Skeele, S. Yoo, Y. Zhang, G. Hollinger, “Multi-UAV
Exploration with Limited Communication and Battery.” IEEE International Conference on Robotics and Automation (ICRA), 2015.
43
![Page 44: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/44.jpg)
Outline
Sensing Coordinating Planning
44
![Page 45: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/45.jpg)
RelatedWork
Sensing Coordinating Planning
45
● “Sampling-based Algorithms for Optimal Motion Planning,” Karaman & Frazzoli 2011.
● “The Stochastic Motion Roadmap: A Sampling Framework for Planning
with Markov Motion Uncertainty,” Alterovitz, et al. 2007. ● “Planning Most-Likely Paths From Overhead Imagery,” Murphy &
Newman, 2010.
![Page 46: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/46.jpg)
PathPlanning
● Trajectory optimization
● Waypoint navigation
● Graph based planning
● Discrete and sampling based planners
46
*Taken from wikipedia.com
![Page 47: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/47.jpg)
Motivation
● Representing the world perfectly is impossible
● Graphs are a versatile representation of many domains
● Making reliable decisions is vital to future of robotics
47
![Page 48: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/48.jpg)
Algorithm:Risk-AwareGraphSearch(RAGS)
● Represent graphs using normal distributions of edge costs
● Search through graph for paths to goal
● Traverse the graph along the path of least risk
(Mean, Variance)
48
![Page 49: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/49.jpg)
● Red path represents A* planning over mean
● Blue paths represents RAGS
− RAGS trades off the lower mean of Red against the path options of Blue.
Execution
(Mean, Variance)
49
![Page 50: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/50.jpg)
● Red path represents A* planning over mean
● Blue paths represents RAGS
− RAGS trades off the lower mean of Red against the path options of Blue.
Execution
(Mean, Variance)
Probability of Better Path At Vs -> A vs B = 41.85% At B -> Vg vs C = 29.62% At C -> D vs Vg = 47.16%
50
![Page 51: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/51.jpg)
QuantifyingRisk
● Current location has neighbor vertices
● Each vertex has child paths to the goal
● Integrate probabilities (of cost) over all child paths
51
![Page 52: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/52.jpg)
QuantifyingRisk
● Probability that the lowest-cost path in the set A is cheaper than the lowest-
cost path in the set B
− Becomes a relationship between mean, variance, and number of paths
− Pairwise comparison of two neighbors
− Provides local ordering
52
![Page 53: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/53.jpg)
Bounding
● Paths with both worse mean and variance are ‘dominated’
● Bounding dominated paths reduces the computational complexity
● Partial ordering
− Only non-dominated nodes are expanded
53
![Page 54: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/54.jpg)
SimulationSetup
Randomly generated graphs
● Final edge costs sampled from edge distributions
● Search from (0,0) to (100,100)
● Compared against A*, D*, and Greedy
54
Edge variances are represented in grayscale
![Page 55: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/55.jpg)
Example
The video shows
1. Generating a PRM (with edge means and variances)
2. Pruning the edges for non- dominated paths
3. Traversing the graph with risk- aware planning
55
![Page 56: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/56.jpg)
SimulationResults
56
Low
er is
better
![Page 57: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/57.jpg)
SimulationResults
57
Low
er is
better
![Page 58: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/58.jpg)
SimulationResults
58
High cost paths!
Low
er is
better
![Page 59: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/59.jpg)
SimulationResults
59
Low
er is
better
![Page 60: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/60.jpg)
Experiments
● Dataset of 64 images
− tree clusters
− man made structures
− varying resolutions.
● Filtered to extract obstacles
● Edge variances taken from pixel intensities between vertices
● Mean values are Euclidean distance
Satellite images for ground robot or low flying UAV
60
![Page 61: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/61.jpg)
Experiments
● Dataset of 64 images
− tree clusters
− man made structures
− varying resolutions.
● Filtered to extract obstacles
● Edge variances taken from pixel intensities between vertices
● Mean values are Euclidean distance plus pixel intensity
Satellite images for ground robot or low flying UAV
61
µ = dist + ave(ip ) '2 = var(ip )
![Page 62: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/62.jpg)
Experiments-Results
62
![Page 63: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/63.jpg)
63
Example:EmptyField
Three distinct scenarios for analysis
Similar trajectories through empty field.
Empty Field
![Page 64: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/64.jpg)
64
Example:SparseTreeCluster
Sparse Tree Cluster
RAGS cuts through sparse cluster to take advantage of open space.
![Page 65: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/65.jpg)
65
Example:DenseObstacle
Large Dense Obstacle
RAGS avoids narrow unlikely path through center of obstacle.
![Page 66: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/66.jpg)
Discussion
● Incorporating uncertainty accounts for unknowns in the real world
● Risk-aware planning provides robustness
● Traditional search methods plan over mean cost risk outliers ● Publication:
○ R. Skeele, J. Chung, G. Hollinger, “Risk-Aware Graph Search”. IEEE International Conference on Robotics and Automation. Workshop on Beyond Geometric Constraints, 2015.
○ Submission planned: Workshop on the Algorithmic Foundations of Robotics (WAFR), 2016.
66
![Page 67: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/67.jpg)
SummaryofContributions
Sensing Coordinating Planning
67
● Monitored dynamic points of
interest
● Leveraged realistic wildfire
simulator for planning
● Demonstrated capability on
hardware
● Introduced coordination method
for uncertain communication
● Simulated large teams of UAVs
cooperatively exploring
● Developed low cost indoor
autonomous quadcopters
● Proposed risk-aware
planning over uncertain costs
● Outperformed traditional
search algorithms
● Demonstrated on satellite
imagery
![Page 68: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/68.jpg)
Sensing Coordinating Planning
SummaryofContributions
68
![Page 69: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/69.jpg)
FutureWork
● Gaussian process model of the fire frontier
○ Would give a continuous model of uncertainty
● Incorporate geometric knowledge of the environment to predict
reconnection
○ Inference techniques on environment structure
● Informative path planning for RAGS
69
![Page 70: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/70.jpg)
Acknowledgements
70
![Page 71: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/71.jpg)
Acknowledgements
Friends, Family
Special Thanks to:
Jen Jen Chung, Carrie Rebhuhn and the entire RDML
71
![Page 72: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/72.jpg)
ThesisCommittee
Advisor Geoff Hollinger
Robotic Decision Making Laboratory
72
Committee Kagan Tumer
Autonomous Agents and
Distributed Intelligence Lab
Committee Bill Smart
Personal Robotics Group
GCR Fuxin Li
Artificial Intelligence, Machine
Learning, and Data Science
![Page 73: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/73.jpg)
Questions?
73
![Page 74: Planning Under Uncertainty for Unmanned Aerial Vehicles (UAVs)research.engr.oregonstate.edu/rdml/sites/research... · The Challenges Precision sensing ability, and/or robust planning](https://reader036.fdocuments.net/reader036/viewer/2022071015/5fce28a7b787914a096a556e/html5/thumbnails/74.jpg)
RelatedWork
Sensing Coordinating Planning
74
● “An Unmanned Aircraft System for Automatic Forest Fire Monitoring and Measurement,” Merino, et al. 2012.
● “Planning periodic persistent monitoring
trajectories for sensing robots in gaussian random fields,” Lan & Schwager, 2013.
● “Sampling-based Algorithms for Optimal Motion Planning,” Karaman & Frazzoli 2011.
● “The Stochastic Motion Roadmap: A
Sampling Framework for Planning with Markov Motion Uncertainty,” Alterovitz, et al. 2007.
● “Planning Most-Likely Paths From
Overhead Imagery,” Murphy & Newman, 2010.
● “Multi-Robot Coordination with Periodic Connectivity,” Hollinger & Singh, 2010.
● “The Sensor-based Random Graph Method
for Cooperative Robot Exploration,” Franchi et al., 2009.