|
To achieve collision free path planning and tracking control of unmanned launch platforms in complex environments, this paper proposes a path obstacle avoidance control method based on deep deterministic strategy gradient reinforcement learning (RL) algorithm. This method models the collision avoidance problem as a Markov decision process, and uses a deep neural network to establish a nonlinear mapping from the laser radar perception state to the optimal control variables of velocity and angular velocity. Compared with traditional path obstacle avoidance algorithms, RL based obstacle avoidance algorithms do not rely on experience to set corner control rules. They can achieve end-to-end mapping from environmental state to optimal control by setting the return function, overcoming the shortcomings of traditional obstacle avoidance algorithms such as weak adaptability to the environment and low generalization ability. The effectiveness and feasibility of the proposed path collision avoidance algorithm were verified through simulation experiments. The results show that compared with traditional path planning obstacle avoidance algorithms, the RL based path avoidance method can achieve obstacle avoidance control in complex environments.
|
