Implementation Of Omni-directional Walking And High-level Decision For Humanoid Robots In RoboCup3D Simulation System

Researches on humanoid robots started in the 70s. Since Honda launches humanoid robot P2 in 1996, the research and development on humanoid robots have become an important field of robotics research so far in the ascendant.Walking of biped robot has been a hotspot of robotics research. The difficulties are reflected in the complexity of implementation, as well as the flexibility and stability of walking. The kinematics model is usually an important prerequisite for actions of robot, especially the walking. In this paper, implementation of forward kinematics was described first in detail according to a specific robot model. Analytic method for solving inverse kinematics was expatiated on. Also a new method using BP neural network to solve inverse kinematics was introduced.Omni-directional walking is humanoid and highly flexible. Key-frame plays an important role in robot walking control system. By break down the omni-directional gait, it can be seen as an integration of the three independent movements. And only planning for the key-frame of each movement is needed. Through a series of comparison, cubic spline interpolation is used to generate other frames while walking through a series of comparison. In this paper, each motion was planned in detail by stages. Tweenframes were generated after key-frames were gotten. And then inverse kinematics was used for control the angle of each joint, thus achieving an omni-directional gait.We have implemented the omni-directional gaits in RoboCup3D simulation platform (SimSpark). Then both BP neural network and reinforcement learning method are used to learn and optimum the parameters for omni-directional walking. Finally the walking gaits can also be flexible,stable and less time–consuming. It can meet the requirements for dynamic target tracking.SimSpark is also an important platform to research on multi-agent system. We have developed a simulation football team named Engine3D in this platform, and research humanoid robots and MAS under the platform. The location method based on communication and credibility, and multi-robot cooperation method of attack and defense base on role switching were proposed. Then with the help of omni-directional walking gaits generator, biped robots are more flexible in doing actions, and more intelligent in high-level decision.