Omnidirectional mobile robots navigation: A joint approach combining reinforcement learning and knowledge-based systems

摘要

A joint approach combining Reinforcement Learning, Knowledge Based Systems and a new methodology for environment mapping around the mobile robot neighborhood, for mobile robot navigation is presented in this paper. The new approach allows cognitive agent based on knowledge based systems, to uses Q-Iearning algorithms to interact successively with a specific domain, in a simulated environment, and once achieved the optimal policy, codes this optimal policy into a symbolic knowledge base that uses first order logic as knowledge representation formalism. Therefore, the knowledge base able to create free collision paths combining basic behaviors is embedded on a omnidirectional mobile robot. The paper also presents the experimental results, with real robots, obtained using the new methodology to capacity the omnidirectional mobile robot to navigate in a dynamic environments. The robot soccer, under the F180 category from RoboCup Federation, is used as an experimental domain.