Hierarchical Game-Theoretic Learning for Mixed Human-Autonomous Vehicle Networks

Project begins 2024.

This project addresses an important need in mixed human/autonomous vehicle teaming. In particular, this work focuses on the development of methods to improve cooperative interactions within an unstructured and unknown environment and enhance dynamic decision-making in heterogeneous teams of humans and autonomous vehicles.

The premise of this research is that a critical human-autonomy interaction exists at both an intravehicle level (i.e., between a human operator and (semi-)autonomous vehicle within the vehicle itself) and inter-vehicle/multi-agent network level (i.e., between multiple autonomous vehicles and human soldiers – including collaborators and adversaries). Both aforementioned levels of human-autonomy interaction are evidenced through separate bodies of literature, with separate, limited adaptation and control and learning algorithms applied for dealing with the interactions at each level. However, the reality is that vehicle-level adaptation and network-level adaptation are coupled; an autonomous vehicle that adjusts its behavior according to the characteristics of the human decision maker or environment represents a time-varying agent whose dynamic time-varying behavior must be accounted for at the network level in order to achieve optimal system-level operation.

This research focuses on the development of a hierarchical game-theoretic adaptive control framework to address human-autonomy interactions at vehicle-/network-levels, accounting for coupling between levels.

Fundamental research questions addressed:

RQ1: What is the impact of uncertainty on the convergence properties of a cognitive level estimate in the presence of adaptive human-autonomy interactions and uncertain environments?

RQ2: What role does the estimated cognitive level play in the underlying decision-making framework (e.g. control architecture design, learning approach) and the system’s ability to leverage libraries of past environmental conditions and human cognitive levels, coupled with corresponding system performances?

RQ3: How do adaptation rates of the agents and the environment, at both the vehicle and network-level, drive the decision-making policies and convergence rates of the proposed framework?

#2.19