Multi-Agent Fuzzy Reinforcement Learning for Autonomous Vehicles

This thesis investigates how the evader in a pursuit-evasion differential game can learn its control strategies using the fuzzy actor-critic learning algorithm. The evader learns its control strategies while being chased by a pursuer that is also learning its control strategy in two pursuit-evasion games; the homicidal chauffeur game and the game of two cars. The simulation results presented in this thesis prove that the evader is able to learn its control strategy effectively using only triangular membership functions and only updating its output parameters. When compared with the simulation results from [1], the approach in this thesis saves a significant amount of computation time. This thesis also introduces fuzzy-actor critic learning to an inertial missile guidance problem. The missile solely relies on its own control surfaces to intercept the target. Proportional navigation is one of the techniques in literature that can successfully guide the missile to the target.