Research on adaptive operation of intelligent robotic arms based on SAC algorithm and hierarchical teaching

An adaptive operation model for intelligent robotic arms based on the integration of the Soft Actor–Critic (SAC) algorithm and hierarchical teaching is proposed. Building upon the existing reinforcement learning control framework, this model incorporates hierarchical teaching's high-level task decomposition and low-level action optimization mechanisms, establishing a comprehensive system spanning state modeling, reward function design, and policy execution. Experimental validation demonstrates that Hierarchical Demonstration Soft Actor–Critic (HD-SAC) achieves a 94.6% success rate in grasping and assembly tasks—approximately 11% higher than traditional SAC—while reducing energy consumption to 111.8 J and completion time to 16.2 s. Even under perturbed conditions, it maintains a success rate above 85% (with the highest reaching 89.3%) and lowers the collision rate to 4.3%, significantly outperforming baseline methods such as SAC and Twin Delayed Deep Deterministic Policy Gradient (TD3). The results confirm that this method exhibits superior stability and robustness in complex tasks.