Research And Implementation Of Path Planning For Mobile Robot Based On Reinforcement Learning

Mobile robot path planning of robot intelligence research is one of the hottest areas. Now the robot’s autonomous learning ability has become the trend of technology development. This paper studies the method of machine learning called Q Learning which is the popular machine learning method. The method does not require the precise environment model and can deal with uncertain situation in the environment.The optimal path is defined in this paper. Mobile robot can find the path by learning "trial and error" in an unknown environment. The path is from start point to goal point to avoid obstacles and minimize the total length. In this paper, an idea is presented with a machine-learning approach to robot path planning, after analyzing the current technology of path planning development.Since Q Learning has the difficult problems of State-space continuity and Action-space continuity, a RBF-Q Learning-GD algorithm framework is proposed. The framework combines RBF network and gradient declined method(GD) which has strong capacity of looking for extreme in unknown distribution, using RBF network approximate Q evaluation function, and using gradient declined method seeking next maximum Q value action. The framework has good generalization.Finally, avoidance model of machine learning is proposed for using Q Learning in an unknown environment, by reasonable definitions of State-space and Action-space, combined with design callback-functions, applying RBF-Q Learning-GD algorithm framework to achieve mobile robot autonomous learning and map out the optimal path. Experimental results show that the designed frameworks can be effectively applied to path planning model, which can solve the difficult problems of State-space continuity and Action-space continuity and can solve similar problems.