基于深度强化学习的协作机器人抓取方法研究

Research on collaborative robot grasping method based on deep reinforcement learning

基于深度强化学习中深度策略性梯度(DDPG)算法,提出加权采样DDPG算法,在仿真环境下,为避免算法探索效率低的问题,将视觉目标检测与加权采样DDPG算法相结合,让目标检测算法获取目标物体所在区域作为算法的先验,使算法探索阶段能够获得更多有效的学习样本,进而提高算法的学习速率和抓取精度。

Based on the deep reinforcement learning algorithm known as deep deterministic policy gradient(DDPG),this study proposes a Weighted Sampling DDPG algorithm for collaborative robot grasping.In a simulated environment,to address the issue of low algorithm exploration efficiency,visual target detection is combined with the weighted sampling DDPG algorithm.This allows the target detection algorithm to obtain the region where the target object is located as prior information for the algorithm,enabling the algorithm's exploration phase to acquire more effective learning samples.and thus improve the learning rate and grasping accuracy of the algorithm.