Compressing the multirobot team formation state based on SOM network

As a platform of multirobots´ cooperation and coordination, the multirobots team formation is paid more and more attention. Using the reinforcement learning to realize the team formation can strengthen not only the self-learning ability but also the self-adaptation. In this research field, however, there still exit problems such as low learning speed and the difficult convergence raised with the exponential space of reinforcement learning. Using the self-organizing map (SOM) network compressing state from exponential to multinomial speeds up the ergodic, consequently improves the learning rate. And the function of adding and deleting the neurons can compress more space. In the simulation of the experiment, the feasibility of these technologies is verified further. The expands of the methods are strong and can be used in the similar system.