Optimal Distribution of Tasks in Human-Autonomy Teaming
Ph.D. Research @ Automotive Research Center
University of Michigan
| Overview
The research created an AI framework to prepare autonomy to collaborate with humans in dynamic strategic operations, considering adversarial dynamics, unforeseen events, human characteristics, workload & resource reduction, and adaptive team design under evolving situations.
A UE4 game platform was developed for actual human players to interact with autonomy in Virtual Reality, with the help of NN-based workload adaptive UI.
Another line of work focused on 1) quantifying and understanding agent- and team-level coordination in dynamically varying situations through causal inference and community detection techniques and 2) adaptation & learning in open novelty.
| Formulation
Heterogeneous Teaming with Load Management Decentralized POMDP
Goal: maximize efficiency with no unnecessary task assignment
Environment: stochastic, uncertain, no centralized coordinator
State: various tasks with dynamic demand levels at multiple locations
Agents: heterogeneous capabilities and decision models
Actions: to execute a task or to idle
Information: local observations and limited communications
Rewards: task completion, load management, individual preference
| Training
Each agent is equipped with a Decentralized Deep Q-Network with bayesian reasoning on past experience and communicated information.
Instead of training with humans, artificial human agents are created with the following characteristics:
Risk takers - noisy rational decisions
Risk tolerance - reward generosity
Adaptation - balancing knowledge and recent observations
Deep Q-Network with Beliefs
| Applications & Studies |
Emergent Strategy Evaluation with Causal Inference and Sub-Team Identification
- (Left) Rule, (Right) RL-trained
Adaptive Team Design
Human-autonomy teaming in Virtual Reality
Cognitive Task Load Adaptive User Interface
Heterogeneity and Risk-aversion
Load Management and Resource Reduction
H. Wu, A. Ghadami, A. E. Bayrak, J. M. Smereka, and B. I. Epureanu, "Evaluating Emergent Coordination in Multi-Agent Task Allocation through Causal Inference and Sub-Team Identification," in IEEE Robotics and Automation Letters, vol. 8, no. 2, pp. 728-735, Feb. 2023. [Open Access]
H. Wu, C. C. Folks, A. E. Bayrak, J. M. Smereka, and B. I. Epureanu, "Human-Autonomy Teaming in Immersive Environments," 2022 Interservice/Industry Training, Simulation, and Education Conference (I/ITSEC), 2022, Orlando, FL. [Full Text]
H. Wu, A. Ghadami, A. E. Bayrak, J. M. Smereka, and B. I. Epureanu, "Task Allocation with Load Management in Multi-Agent Teams," 2022 International Conference on Robotics and Automation (ICRA), 2022, pp. 8823-8830, Philadelphia, PA. [Preprint Version]
--------, "Impact of Heterogeneity and Risk Aversion on Task Allocation in Multi-Agent Teams," in IEEE Robotics and Automation Letters, vol. 6, no. 4, pp. 7065-7072, Oct. 2021. [Presented Virtually at 2021 IROS, Prague, Czech Republic] [Open Access]