Experiments in HumanRobot Teams

1
Experiments in Human-Robot Teams

• Curtis W. Nielsen, Michael A. Goodrich, Jacob W.
  Crandall
• Brigham Young University

2
Motivation

• Search and Rescue Robotics
• Still in its infancy
• Current methods have very high workload

3
The Questions
How do human-robot interactions affect team
performance and human workload? Where is the
Sweet Spot?
4
Procedure

• Domain
• Topological map-building
• Interaction Schemes
• Teleoperate
• Point to Point
• Region of Interest
• Experiment

5
Behavior-based Landmarks

• Set of behaviors afforded to the robot
• Affordance the perceived actionable properties
  between the world and an actor (Gibson)
• Actor robot
• Afforded behaviors turn right, turn left, go
  forward
• Afforded behaviors are found using filtered sonar
  measurements

6
Building a Topological Map
Classify a landmark
Disambiguate landmarks
Choose an action
7
Characterizing the interaction schemes

• Landmark classification
• Landmark disambiguation
• Choose an action
• Advantages
• Disadvantages

8
Teleoperate (TOL)

• Choose an action Human
• Landmark classification Human
• Landmark disambiguation Human
• Advantage Human has very high control of the
  movement of the robot
• Disadvantage The human must devote a lot of
  attention to the robot

9
Point To Point (PTP)

• Choose an action Human
• Landmark classification Robot
• Landmark disambiguation Human
• Advantage Relatively low workload
• Disadvantage Requires human control for each new
  action

10
Region of Interest (ROI)

• Choose an action Human / Robot
• Landmark classification Robot
• Landmark disambiguation Robot
• Advantage Very little human workload
• Disadvantage Takes a long time to disambiguate
  landmarks

11
The interface
12
Joystick Control
Action Selection
Landmark Classification
Landmark Disambiguation
13
Point to Point Control
Action Selection
Landmark Classification
Landmark Disambiguation
14
Region of Interest Control
Action Selection
Landmark Recognition
Landmark Disambiguation
15
Measuring Performance
The time it takes for the system to complete an
accurate map of the environment.
Time
16
Measuring Workload Behavioral Entropy

• Entropy of the joystick (Boer)
• Velocity of the mouse.
• Button clicks on the mouse and joystick
• Change robots
• Scaling issues

17
Experiment 10 subjects
18
Region of Interest
19
Point to Point
20
Mixed with Joystick
21
Workload(without joystick)
22
Elapsed Time (without joystick)
23
Workload (with Joystick)
24
Elapsed Time (with Joystick)
25
Results
Tradeoff Curve
With Teleop
Without Teleop
26
Conclusions

• Measured performance and workload for a system
  where a human controls 3 robots in a map-building
  task.
• Analyzed the tradeoffs in terms of workload and
  performance of changing interaction schemes
  between robots.
• Found a sweet spot where performance is
  relatively high and workload is relatively low.
• Sweet spot can change as representation and
  autonomy level change.

27
Questions for Future Work

• Vary the number of robots?
• Vary the number of users?
• Vary environment complexity?
• Dynamic autonomy?
• Workload measurements (scaling issues)?