Study On Target Recognition And Strategies Planning Algorithm For Soccer Robot Systems

Multi-robot system is a popular issue in the field of robotic researches, which includes multi-robot cooperation, task allocation, communication and path planning. Compared to single-robot system, it shows obvious advantages on the scope of application, ability of adaption, efficiency and quality of task completion. With the development of modern technology and increase of economic demand, researches of multi-robot system has advanced in an enormous pace especially on building models in a complicated and practical environment.Robot soccer game has been developed into a technology competition in recent years, which involves a number of fields such as robotics, perception and fusion, communication, vision and image processing, machine learning, reasoning and decision making and so on. As a typical multi-intelligent-agent cooperation system, it provides us an excellent platform for experiments.Using small-size robot system based on global vision technique, this study mainly focuses on the problems of target recognition, path planning and dynamic task allocation in a complicated environment with an uncertain target. Progress has been made by this study in the following aspects:1. To reduce the influence to robots’ decision caused by mutative brightness in the game site, a dynamic template model is proposed which takes account the effect of time-varying data. Combined with the traditional clustering data reduction model, it improves SVM (Support Vector Machine) sorting algorithm to achieve the stability and real-time performance in mutative brightness by introducing a time-control factor. Besides, the dataset is built by interval data extraction to simulate the brightness in rapid change. The stability and efficiency of this model has been testified by using standard KDDCUP1999dataset.2. To increase the efficiency of task execution in an environment with a mobile target, a fast task assignment algorithm based on Voronoi graphs is proposed. Considering the characteristics of task assignment in actual situation, it adds new measures such as mission execution supervision and over waiting prevention to the traditional auction to solve the dead lock problem. And the termination of auction is triggered by the single mission to reduce the rate of dead lock and to improve the robustness in the mission assignment process. 3. To solve the path planning problem of mobile robots in the dynamic complicated environment, a rolling window method based on obstacle prediction is proposed. This method combines the advantages of static global path planning algorithm and dynamic local path planning together effectively. In local level, it uses the rolling window to plan the local path on the basis of predicting the movement of obstacles. In global level, the main direction search strategy is used in the static global path planning algorithm to select the prior path in the rolling window using the direction consistency. This method achieves both effects of avoiding obstacles in the global level and avoiding collision in the local level, which improves the efficiency of the robot decision system in dynamic path planning.Finally, we run experiments on the small-size robot system in different formation in the environment with varying brightness and glare interference. The result shows that three strategies presented here perform well in robot decision system and improve the decision ability effectively.