Hybrid Reinforcement Learning and Uneven Generalization of Learning Space Method for Robot Obstacle Avoidance

摘要

This paper introduces a hybrid reinforcement learning algorithm for robot obstacle avoidance. This algorithm is based on SARSA (λ), and mix with the supervised learning. This hybrid learning algorithm can reduce the learning time obviously which is demonstrated by the simulations. In reinforcement learning, generalization of learning space is important for learning efficiency. An uneven generalization model is designed for improving the learning efficiency. The simulations show that the uneven model can not only reduce the learning time, but also the moving steps.