Modeling Collision Avoidance Decisions by a Simulated Human-Autonomy Team
Report Number:
ARL-TR-9626
January 10, 2023
Approved for public release: distribution is unlimited.
Author(s):
Evan C Carter and Vernon J Lawhern
Abstract:Creating autonomous teammates that respond to their human counterparts in an adaptive way requires models of human goal and intent. One approach to building such models is inverse reinforcement learning, a process by which a reward function, representing goals and intent in a task, is learned from observed behavior. Here, we validate an approach to inverse reinforcement learning in a collision avoidance task completed by a simulated human-autonomy team. Our findings indicate several promising aspects of our approach, as well as some clear drawbacks for future work to investigate.
