Abstract
Reinforcement learning is a learning framework that is especially suited for obstacle avoidance and navigation of autonomous mobile robots, because supervised signals, hardly available in the real world, can be dispensed with. We have to determine, however, the values of parameters in reinforcement learning without prior information. In the present paper, we propose to use a genetic algorithm with inheritance for their optimization. We succeed in decreasing the average number of actions needed to reach a given goal by about 10-40% compared with reinforcement learning with non-optimal parameters, and in obtaining a nearly shortest path.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have