Intelligent control and learning techniques for mobile robots.

摘要

This dissertation presents intelligent control and learning techniques for mobile robots. Two issues in intelligent control techniques are addressed: the intelligent control techniques for point-to-point navigation and behavior learning. A two-level architecture was implemented using the multi-agent robot control architecture called the IMA. Under the IMA, all hardware interfaces are designed as the basic components, represented as the IMA resource agents. The basic behaviors are constructed in both low-level and high-level control layers using two designing methods: potential field and fuzzy control.; Behavior learning was based on the sensory-motor coordination technique. The correlation technique was used to analyze the data sets on test runs. The results were used for designing a behavior learning system using fuzzy logic and implemented in the low-level control layer.{09}Reinforcement learning was applied to correct the fuzzy rules for the learn-to-avoid-obstacle behavior. Results indicate a good learning performance in dynamic environments.