Research on Multi Robot Path Planning based on Improved DQN

A common and important problem in the deep reinforcement learning algorithm DQN (Deep Q-learning Network) for multi robot path planning is low sample efficiency. To solve this problem, this paper proposes a SSDD-DQN (Simple and Space based Dueling Double Deep Q-learning Network) method for simple and sparse intelligent factory environments. Firstly, this method uses a priority based post experience replay strategy based on chaotic sequences to optimize the later training process of the model; Secondly, an improved state space balance has been introduced <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$E$</tex> -Greedy exploration action selection strategy to deal with the equilibrium problem of exploration and utilization of intelligent agents; Finally, a consensus based congestion avoidance mechanism inspired by social consciousness was introduced. The experiment shows that this method has high sample efficiency, low solution mobility consumption, and can significantly alleviate congestion among multiple robots.