Research on Collision-Free Control and Simulation of Single-Agent Based on An Improved DDPG Algorithm

This technical note investigates a novel design of algorithmic models for single-agent obstacle avoidance based on deep reinforcement learning. In view of the traditional DDPG algorithm has many shortages such as many training rounds and requiring much time to learn. To this end, an improved deep deterministic policy gradient algorithm is proposed and used to improve the efficiency of obtaining obstacle avoidance path. In other words, the proposed strategy may be favor for overcoming the limitations of traditional DDPG algorithm by adopting two-way search of agents, improving the reward mechanism and adding real-time rewards and punishments as supplements. Different from the traditional human strategy obstacle avoidance algorithms such as artificial potential field method and grid method, the bidirectional DDPG algorithm has the abilities of independent decision-making, stronger adaptability and better obstacle avoidance effect in many practical applications.