Hybrid Reinforcement Learning in Autonomous Mobile Systems

S.M. Youssef (Egypt)


Learning Robots, Motion Planning and Learning, Artificial Intelligence, Autonomous Mobile Agents.


. We introduce a mechanism for learning in mobile autonomous systems. It is difficult to apply traditional reinforcement learning algorithms to mobile agents, due to problems with large and continuous domains, partial observability, and limited number of learning experiences. This paper deals with these problems by combining hierarchical reinforcement learning, based on subgoal discovery, with memory, implemented using a short-term memory feedback neural network whose inputs are the discrete events extracted from abstract observations which cluster raw input data. Experiments show that this method outperforms flat reinforcement learning methods and demonstrate the methodology’s feasibility.

Important Links:

Go Back