Synchronous vs. Asynchronous Video in Multi-Robot Search

1
Synchronous vs. Asynchronous Video in Multi-Robot
Search

• Prasanna Velagapudi, Jijun Wang, Huadong Wang,
• Paul Scerri, Michael Lewis, Katia Sycara
• University of Pittsburgh
• Carnegie Mellon University

2
Urban Search and Rescue (USAR)

• Location and rescue of people in a structural
  collapse
• Urban disasters
• Landslides
• Earthquakes
• Terrorism

Credit NIST
3
USAR Robots

• Robots can help
• Unstable voids
• Mapping/clearing
• Want them to be
• Small
• Cheap
• Plentiful

Credit NIST
4
Urban Search and Rescue (USAR)

• Now One operator ? one robot
• Directly teleoperated
• Victim detection through synchronous video
• Future One operator ? many robots
• Manufacturing robots is easy
• Training operators is hard
• Need to scale navigation and search

5
Synchronous Video

• Most common form of camera teleoperation
• High bandwidth
• Low latency
• Applications
• Surveillance
• Bomb disposal
• Inspection

iRobot PackBot
6
Synchronous Video

• Does not scale with team size

7
Synchronous Video

• Does not scale with team size

8
Synchronous Video

• Does not scale with team size

9
Asynchronous Imagery

• Inspired by planetary robotic solutions
• Limited bandwidth
• High latency
• Multiple photographs from single location
• Maximizes coverage
• Can be mapped to virtual pan-tilt-zoom camera

10
Hypothesis

• Asynchronicity may improve performance
• Helps guarantee coverage
• Can review images multiple times
• Asynchronicity may reduce mental workload
• Only navigation must be done in real-time
• Search becomes self-paced

11
USARSim

• Based on UnrealEngine2
• High-fidelity physics
• Realistic rendering
• Camera
• Laser scanner (LIDAR)

http//www.sourceforge.net/projects/usarsim
12
MrCSMulti-robot Control System
13
MrCSMulti-robot Control System
Status Window
Map Overview
Video/ Image Viewer
Waypoint Navigation
Teleoperation
14
Experimental Conditions

• Objective
• Find victims ? Mark victims on map
• Control 4 robots
• Waypoint control (primary)
• Direct teleoperation
• Explore the map
• Map generated online w/ Occupancy Grid SLAM
• Simulated laser scanners

15
Experimental Conditions
10 Victims
16
Experimental Conditions

• Streaming Mode

• Panorama Mode

17
Experimental Conditions(Streaming Mode)
18
Experimental Conditions(Panorama Mode)
19
Subjects

• 21 paid participants
• 9 male, 12 female
• No prior experience with robot control
• Frequent computer users 71
• Played computers games gt 1hr/week 28

20
Method

• Written instructions
• 15-20 min. training session
• Both streaming and panoramas enabled
• Encouraged to find and mark a victim
• 20 min. testing session
• 20 min. testing session

21
Metrics

• Switching times
• Number of victims
• Thresholded accuracy

22
Victims Found
Average of victims found
23
Trial Order Interaction
Average of victims found
24
Switching Time (Streaming Mode)
Average of reported victims
25
Switching Time (Panorama Mode)
Average of reported victims
26
Conclusions

• Streaming is better than panoramic
• Perhaps not by as much as expected
• Conditions favorable to streaming video
• Similar asynchronous performance is good
• May avoid forced pace switching
• May scale with team size

27
(No Transcript)
28
Operator-induced latency
Operator switch time
of robots
29
Victims Found

• Repeated Measures ANOVA
• 1.5m radius
• F(1,19) 8.038
• p 0.01
• 2.0m radius
• F(1,19) 9.54
• p 0.006

30
Trial Order Interaction

• Repeated Measures ANOVA
• 1.5m radius
• F(1,19) 7.34
• p 0.014
• 2.0m radius
• F(1,19) 8.77
• p 0.008

31
Switching Time

• Streaming mode
• Repeated Measures ANOVA
• F(1,19) 3.86
• p 0.064
• Panorama mode
• No relation found