Applying Reinforcement Learning to Obstacle Avoidance
Josh Beitelspacher · 2001
This paper applies reinforcement learning techniques to an asteroids-type game. Both Q-Learning and Sarsa(λ) are used to learn obstacle avoidance. Action-value function approximation is provided by a backpropa-gation neural network. The combination of these two methods produces obstacle avoid-ance strategies that perform at nearly a hu-man level. 1.