Autonomous Learning And Navigation Control For Mobile Robots Based On Reinforcement Learning

Mobile robots are a kind of intelligent systems that can move and work autonomously in certain environment, and have been widely used in the area of industry, agriculture, daily life and military affairs. Navigation is the most fundamental and important problem for mobile robot and it is necessary for mobile robot to have the ability of moving around autonomously and safely. While the control methods based on robot learning is the key to achieve autonomous navigation.Among the learning methods, reinforcement learning (RL) has attracted most researchers in the area of robotics. It not only has strong on-line adaptability and self-learning ability for complex system, but also has the merits of human-like thinking mode. But with the development of mobile robot, more challenges come up, such as environment perception, generalization of RL, reactive control in local environment, qualitative navigation based on hybrid control, etc. Hence it is a pivotal research subject to research on robot learning and navigation with the help of the development in artificial intelligence, automatic control and robotics.After the survey and analysis of the current research work, we present in this thesis our research on the problems of mobile robot learning and navigation control, which focuses on reactive control and hierarchical navigation based on RL and the derived methods. Related techniques of information fusion, knowledge representation, learning and control algorithms are proposed according to different navigation missions and environments. Moreover the proposed learning and control algorithms are verified through simulated experiments and real mobile robot. The main contributions of this thesis are:(1) Data fusion and knowledge representation methods based on multi-ultrasonic sensor system are presented.A multi-ultrasonic sensor system is designed for mobile robots. Through sensing the environments with multi-ultrasonic sensor detecting system, the method of target identification and further active detecting for interesting targets is proposed, which achieves exact identification of the indoor environments with special features. In this method, different environment features are classified and a multi-ultrasonic sensor system is used to provide related TOF (Time-of-Flight) information. According to the TOF information, different decisions are made through Dempster-Shafer evidential reasoning and further active detecting.Based on the grey system theory, the concepts of grey measurement system and grey sensors are discussed from the point of view of incomplete information processing compared with numerical and symbolized measurement system. The methods of grey representation and information processing are proposed for data collection and reasoning, which establishes the foundation of hybrid map-building in qualitative navigation and grey control rules-based learning...