1 of 14

A Risk-Aware Leadership Structure for

Heterogeneous Robotic Teams

Mark Allison∗ , Alice Gaspard , Max Parks

Department of Computer Science,

University of Michigan – Flint

Funded by the National Science Foundation Award # 2220513.

2 of 14

Overview

  • Challenges in Scaling Up Large Teams of Heterogeneous Robots
  • Leadership
  • Risk
  • Approach
  • Preliminary results

3 of 14

Problem Domain – Large Scale Heterogeneous Robotic Teams

  • Heterogeneous Teams
    • Complementary capabilities in order to accomplish more complex tasks
    • Higher-dimensional data sets through coordinated perception
  • Large Scale Teams:
    • Increased communication to coordinate
    • Non-linear increase in performance
    • Challenge for Human collaboration

4 of 14

Leadership

​

  • Universal feature of human societies (Brown, 2004)
  • Any social model of collaborating robots would necessarily benefit from a similar structure as exhibited in human social teaming
  • Should participate in decision-making in HRTI
    • = time criticality, who has the skill to decide (human or robot), and cost or severity of consequence (Abbass, 2019)
  • Should persist during mission

5 of 14

Leadership

6 of 14

Recall SLO Architecture

7 of 14

Persisting Leadership

8 of 14

Risk

  • Assessed as the probability of an event occurring times the cost (impact) should it occur
  • Identifying the events, their cost, and contingencies is situational, computationally intensive and non-trivial
    • Will need to be prescribed pre-mission by human (for now).
  • Threshold needed to determine which risks are within the purview of the robot team vs human

9 of 14

Risk Analysis - Fault Tree

10 of 14

Experimentation - Simulation

​

  • RQ1: What is the computational overhead of the leadership structure as the team size scales upwards?
  • RQ2: How does the leadership structure impact communication as the team size scales upwards?

11 of 14

Preliminary Results RQ1

12 of 14

Preliminary Results RQ2

TABLE I: Average Message Dissemination Times

​

Team Size 10 20 30 40 50 60

​

Leaderless 143 259 488 772 1410 2336

​

With Leader 106 156 310 497 801 1070

13 of 14

14 of 14

References

D. E. Brown, “Human universals, human nature & human culture,”

Daedalus, vol. 133, no. 4, pp. 47–54, 2004.

​

H. A. Abbass, “Social integration of artificial intelligence: Functions,

automation allocation logic and human-autonomy trust,” Cognitive Computation, vol. 11, no. 2, pp. 159–171, 2019.

​

​