Human Decisions versus Multi-agent Reinforcement Learning in a Pursuit-evader Game
T. Adams, Andrew C. Cullen, Tansu Alpcan · IFAC-PapersOnLine · 2024
The modelling and analysis of nonlinear cyber-physical systems is integral to applications ranging from social networks to defence strategy. However, conventional linear-quadratic methods inadequately capture the complex nonlinear behaviour of human decision-makers in these systems. Utilising game theory and multi-agent reinforcement learning (MARL), we explore dynamic interactions between humans and machines within a cyber-physical environment, using a swarmalator to represent strategic humans in a pursuit-evader game. Preliminary results indicate that MARL inadequately captures the nonlinear behaviour of human decision-making, with pursuers and evaders exhibiting human behaviours achieving the highest utilities.