Solving Game-Theoretic Hamilton-Jacobi Equations in a Model-Free Way

Kyriakos G Vamvoudakis
Georgia Institute of Technology

The use of reinforcement learning to solve game-theoretic-based control problems is an area that has attracted a significant amount of research attention in recent years and will continue to grow, as autonomy becomes a necessity. This talk will highlight some new methods for the design of optimal game policies that do not require full information of the physics of the systems. Optimization-based design has been responsible for much of the successful performance of engineered systems in aerospace, industrial processes, vehicles, ships, robotics, and elsewhere since the 1960s. We will present novel data-driven methods for nonlinear game-theoretic problems evolving in a continuous-time sense. The dynamics of the players do not need to be known for these online solution techniques. These methods implicitly solve the required game design equations without ever explicitly solving them. Different aspects of the control inputs (decision makers) in terms of cooperation, collaboration, altruistic versus selfish behavior, antagonism, competition, incentives, cheating, and other concepts of multiplayer team play will be explored.

Back to Workshop I: High Dimensional Hamilton-Jacobi Methods in Control and Differential Games