Biomimetic Control And Behavior Evolution Of Quadruped Robot

Legged robot is an active field in the study of robot in recent years. It is a high integration of robotics, biology, control theory, sensor technology, signal processing, etc. Because of its trait which is different from the wheeled robot or derailed robot, scientists from many countries pay much attention to it. Among the legged robots, the quadruped robot is an ideal research object because of the much more simple control method and structure. The quadruped robot can be used not only in military, planet exploration, catastrophic reaction but also in bionics and home entertainment.Sponsored by the National Nature Science Foundation of China and the Innovation Fund projects of the Chinese Academy of Sciences, this thesis focused on the research and development of a quadruped robot system. The whole thesis involves many research fields manifested in following several aspects:1. Made the kinematics and dynamics model based on the study of motion organ and motion philosophy of the quadruped, and developed the quadruped robot successfully.2. Designed a CPG with proprioception. The earlier control method based on the bionics only imitated the basic behavior and function of the creature, and it is effective when it is executing simple and regular tasks. But it is less useful when it comes to a complicated task and environment. With the development of biology and neurology, the new method and concept were brought into the biomimetic control field. We designed a novel CPG with proprioception from the practical engineering facet, and controlled the quadruped robot with it.3. Reflex and dynamic regulation used in neurology are brought into the CPG, with which the quadruped robot was controlled in complicated environments. Reflex is a basic element of the creature behavior. It is the creature's reaction for the outward environment and a mirror to its adaptation ability. The adaptive ability of the robots was improved by bringing reflex into the CPG, so it widens the content of the biomimetic Control method.4. This thesis focuses on the behavior evolution based on the characteristic of the mammal behavior evolution. The reinforcement learning was used and CMAC neural network constructed the reinforcement learning system,implemented the Q-learning. Study the self-navigation in the unknownenvironment considering the characteristic of quadruped robot motion, itcan stride some lower obstacle; Applying the evolutionary learning, weimitated the progress of learning walking of a quadruped. With it,quadruped robot can learn to motion itself. It is important for thedevelopment of the quadruped robot. The ability can be used in otheraspects and it will make the quadruped robot have the learning ability that acreature has. This will improve the emergency action of a quadruped robotand it is more important for a quadruped robot exploring the planet.The original contributions of this thesis: The thesis designed a novel CPG withproprioception from the engineering point based on the development of biology andneurology; The reflex and the dynamic linkage are brought into the CPG and controlthe quadruped robot in complicated environment; Study the behavior evolution ofquadruped robot with reinforcement learning and evolutionary learning based on thecharacteristic of the mammal behavior evolution. It improves the environmentadaptation ability and can get the motion ability itself without any other earlierinformation.