Multi-robot Cooperation Based-on Learning And Evolution

The multi-robot system has been one of the most interesting and challenging problems of robotics research in the recent years. It includes a wide range of research topics and applications, such as multi-robot cooperative transportation, exploration and mapping, robot soccer, etc. Multi-robot system can deal with tasks that are difficult to be accomplished by an individual robot. A team of robots may perform the assigned task in a more reliable, faster, or cheaper way through cooperation. Therefore, an increasing amount of robotics research focuses on multi-robot cooperation. In order to develop ability to cope with unexpected situations, one possible solution to generate cooperative behavior in the multi-robot system is to let the robot learn and evolve through the interaction with the environment and other robots.In this dissertation, cooperative transportation is considered as a task of a multi-robot system, researches mainly addressed on cooperative behavior acquisition based on reinforcement learning and cooperative coevolution. The major contributions of this dissertation are as follows:1. A comprehensive review is given on the classification, main research fields, research approaches and research results of multi-robot system. Investigation methods, primary results, research trends, and main problems of evolutionary robotics applied evolutionary algorithms to evolving robot and multi-robot behaviors are summarized. The unsolved problems of reinforcement learning applied to multi-robot systems are discussed.2. A simulation environment for multi-robot cooperative transportation is designed and implemented. It provides a test-bed for multi-robot cooperation reaearch based on reinforcement learning and evolutionary algorithms.3. A distributed two-layered reinforcement learning approach is presented for acquiring multi-robot cooperative behavior. Simulation experiment that three robots uplift a disk-like object through cooperation is conducted. The effectiveness of the presented approach is validated.4. A cooperative strategy based on action predicting and reinforcement learning is constructed. A predictor is introduced to predict the actions of other robots, robot determines appropriate action based on predicting results. Simulation experiment results that three robotsuplift a disk-like object show that the strategy is successful for multi-robot cooperative behavior acquisition.5. A scheme to evolve multi-robot cooperative behavior based of cooperative coevolutionary algorithm is investigated for multi-robot transporting multiple objects. It combines task allocation with multi-species coevolution. Simulation results shown that the performance is superior to existing methods.6. A modified potential field method for local obstacle avoidance and navigation is presented. It adopts a circular active window, an adaptive goal attractive potential field function, and an enhanced steering direction signal. The proposed approach has been implemented and tested on a mobile robot. Results have shown that the robot is able to successfully perform local navigation with stronger obstacle avoidance power and simple algorithm.