An Improved TD3 Algorithm for 3D Path Planning of Robotic Arm
In the area of military aviation,complicated tasks pose challenges to the path planning of robotic arms.To solve the problems of low learning efficiency and low sample utilization of Twin Delayed Deep Deterministic policy gradient(TD3)algorithm,an improved TD3 algorithm of Recurrent-TD3 is proposed.First-ly,Long Short Term Memory(LSTM)is integrated into strategy network and value network to capture time series information of aviation control tasks,enhance its response ability to time series changes,and enable it to consider historical actions and states in decision-making,and improve the representation ability of the network.Then,Hindsight Experience Replay(HER)is integrated into the TD3 algorithm to avoid the difficulty in learning the sparse rewards in tasks,thereby making more efficient use of the samples by converting the experience of not reaching the goals into the experience of reaching the new goal.Finally,a collision detection process based on the bounding box is designed to improve the safety of robotic armi military aviation missions.The experiments show that this method can find a collision-free path faster than other methods,and the average path length is the shortest.
万方数据
维普
