An Object Recognition Grasping Approach Using Proximal Policy Optimization With YOLOv5

Aiming at the problems of traditional grasping methods for mobile manipulators, such as single application scenarios, low accuracy, and complex tasks, this paper proposes an object recognition approach using Proximal Policy Optimization (PPO) with You Only Look Once v5 (YOLOv5), which combines a vision algorithm deep reinforcement learning to achieve grasping. First, YOLOv5 is adopted identify obtain location information. Second, PPO used strategy. Third, compared Soft Actor-Critic (SAC) Trust Region (TRPO) algorithms in batches 16 128, respectively. The average reward training results PPO, SAC, TRPO are obtained our work. Experimental show that proposed method, terms speed, outperforms original YOLOv4 model. model achieves 96% precision on own built dataset, has higher detection lower hardware requirements than Our method SAC grasping, improved by 93.3% 41% algorithms, Finally, through comparison ablation experiments, highest accuracy mean (mAP)@0.5 value 92.3%. We demonstrate actual physical experiments success rate under reaches 100%, providing new research strategy robot manipulator.