Towards an Understanding of the Impact of Autonomous Path

Towards an Understanding of the
Impact of Autonomous Path Planning
on Victim Search in USAR
Paul Scerri, Prasanna Velagapudi, Katia Sycara, Huadong
Wang, Shih-Yi James Chien and Michael Lewis
Robotics Institute, Carnegie Mellon University
School of Information Sciences, University of Pittsburgh
Urban Search and Rescue
•
Focus on Chemical,
Biological, Radiological,
Nuclear events
•
Use multiple robots to
search for victims in
dangerous urban disaster
environment
•
Complex environment
means that humans are
required
Operator Tasks
•
Operator has several independent tasks
•
Processing vision data (identifying victims/problems)
•
Rescuing stuck or broken robots (trapped under
chairs or high centered)
•
Planning exploration
•
Coordinating robots
Increasing Robot:Operator
Ratio
•
Operators are extremely expensive compared to robots
•
Easier to get more robots than more operators
•
Robots spend much of their time moving slowly
between locations
•
Operator time is not efficiently utilized
•
Unpredictably required
Autonomous Path Planning
•
Robots are already performing SLAM w/ LIDAR data
•
Allow robots to plan their own paths, to cooperatively
explore the environment
•
Path planning is mature, reasonably reliable in some
environments
•
Suspect that operators spend a lot of effort thinking
about path planning, for little gain
Tradeoffs of Autonomy
•
Large amount of operator time saved, corresponding
increase in efficiency
•
Robots use abstracted data to decide where to explore,
human insight/semantic knowledge might be more
efficient
•
Operators may lose some situational awareness if they
don’t need to control robots
Lattice Planning
•
Straightforward implementation of published algorithms
•
Nodes valued by expected information gain of going to that
location
•
Edges valued by probability of traversing safely
•
•
Thresholded, with bias against paths of other robots
Branch and bound search to find path that maximizes
information gain
•
Some limits on path length, preference for straightness, etc.
Nodes
Edges
Occupancy grid
Path
USARSim
•
High-fidelity simulator
based on Unreal
Tournament
•
Real-time physics with
physics card
•
Open source, freely
available
•
Maintained by NIST
Experiment Design
•
60 paid subjects in 30 teams of 2 used both designs
•
24 P3ATs, 25 min., large office environment, find/mark
victims
•
Auto:
•
•
Path planner, with operator able to teleop or waypoint plan
Manual:
•
Waypoint planning for each robot, teleop when required
Thumbnails for
each robot
Enlarged video and teleop
Map and Victim Marking
More Area Explored
(p < 0.001)
Auto
Manual
(Manual condition: 4.26 robots ignored)
More Victims Found
(p = 0.003)
Auto
Manual
(Same victims/area in both groups)
Decreased Marking Error
p = 0.002
Auto
Manual
Same Workload
Auto
Manual
Conclusions
•
Autonomous path planning a useful way of reducing
operator load when environment allows it
•
•
Operator’s time taken up with other activities
•
•
Benefits (faster planning, handling more robots)
outweigh costs (loss of situation awareness, lack of
human insight)
Not clear they fully exploit all robots
Future focus on data presentation/visualization