A FRAMEWORK FOR AUTONOMOUS VEHICLES WITH GOAL INFERENCE AND TASK ALLOCATION CAPABILITIES TO SUPPORT PEER COLLABORATION WITH HUMAN AGENTS

摘要

This paper proposes a framework for autonomous vehicles to collaborate with human agents as peers in task completion scenarios. In this framework, the autonomous vehicles utilize the Bayesian inference method to determine human intention. An optimal task allocation that minimizes the mission completion time while respecting the intention of the human agents is developed using the Mixed Integer Linear Programming (MILP) method. The proposed framework can accommodate different levels of suboptimality in human agents' behavior by adjusting a tunable parameter in the inference model. The effectiveness of the framework in facilitating human-autonomous vehicle collaboration is demonstrated through simulations.