Research And Realization Of The Architecture And The Basic Ability Of The Robocup 3D Robot

The ultimate goal of the Robocup Organization is to develop a team of fully autonomous humanoid robots that can defeat the human world champion team in soccer by 2050. Robocup has great significance to promote industry, academia and research. The AmoiensisNQ robot soccer team of Xiamen University is led by the author of this article. In recent years, it has taken part in several competitions at home and abroad, and has achieved favorable results. The team has made some breakthroughs in various areas of theoretical research. This paper summarizes the research of this team over recent years.Robocup 3D is a competition that is based on humanoid robot simulation. The simulation is based on the software "SPADES", which is an abbreviation for "Simulation of Parallel Agent Discrete Events" It is an abstract time sequenced simulation in continuous space. Robocup 3D uses publicly recognized ODE as the basis of the physical simulation engine, making the system more realistic for imitating the real world.A Robocup 3D robot is the focus of our study. This robot, according to the characteristics of the publicly supplied "self-reconstruction robot system", is constructed using the "layered structure model". Software agents of the robot include communication, skill, and decision-making layers in this structure. In order to satisfy the constraints imposed on the robot, e.g. limited calculation time, we used a multi-threaded asynchronous approach to resolve time conflicts between updating the robot world model and the time taken with decision-making.The basic skill of the robot is vision based target tracing of the ball and other robots in the playground. In this paper, we use the ODEengine "simulation conditions and prediction" algorithm to track the ball, based upon its movement. We implemented our "ball chasing" abilities by using the analytic method suggested by Peter.The "walking" of the robot is the focus of Robocup 3D, and is also important for general bipedal robot study. "Step planning" is the most important component for the "walking" of the robot. Generally speaking, step planning includes such methods as : - "reference path" method, "walk data" method, central pattern generation, and other methods. The stability of the walking involves both "dynamics equations" and the ZMP theory. This paper describes the general principle of the ZMP theory. Applying this theory to the structure of the Robocup robot, we plan the stepping and tracking of the robot. In order to lower the on-line simulation time, we created an "automatic walking adjustment" algorithm and achieved better results.On the basis of robot walking planning, in order to realize the adaptive walk of robot, the paper proposes that using genetic neural networks to evolve an adaptive control neural network. In order to improve the efficiency of evolution, we use the neural network which data structure can be simplified; By incentive mechanisms that encouraging the reduction of the number of neurons in the evolution, we speed up the evolution of computing. By using the abnormal samples of Robot walk and creating a suitable fitness equation to train the neural network, we obtain a better walk results finally, and realize the adaptive walk of robot.