Study on learning algorithm of transfer reinforcement for multi-agent formation control

Considering the obstacle avoidance and collision for multi-agent cooperative formation in multi-obstacle environment, a control algorithm based on transfer learning reinforcement is proposed. Firstly, source task stage, large storage space required by Q-table solution avoided using value function approximation method, which effectively reduces requirement improves solving speed of algorithm. Secondly, phase target task, Gaussian clustering was used to classify tasks. According distance between center optimal class selected learning, negative phenomenon, improved generalization ability convergence Finally, simulation results show that this method can form maintain configuration system complex environment with obstacles, realize at same time.