طراحي كنترل‌كننده تناسبي-انتگرالي-مشتقي مبتني بر يادگيري تقويتي عميق براي يك نمونه ربات پوشيدني

This research presents a novel method for designing an intelligent Proportional-Integral-Derivative (PID) controller for a 3-DOF wearable robot performing the sit-to-stan‎d (STS) rehabilitation task. The main objective is to automate an‎d optimize the controller gains online using artificial intelligence, overcoming the challenges of controlling nonlinear an‎d Multi-Input Multi-Output (MIMO) systems. To this end, a control architecture based on feedback linearization was chosen as the underlying structure, an‎d two families of intelligent methods were implemented an‎d compared for tuning the PID coefficients: four Deep Reinforcement Learning (DRL) algorithms (DDPG, TD3, SAC, PPO) an‎d two Spiking Neural Network (SNN) models (Rate-Based an‎d LIF). To ensure a fair comparison an‎d achieve optimal performance, the hyperparameters of all algorithms were systematically optimized using the Taguchi method. Simulation results in the MATLAB environment demonstrate that all intelligent controllers significantly outperform the classic PID. A more detailed comparative analysis revealed that the SNN-based controllers, particularly the LIF model, exhibited superior performance compared to the reinforcement learning algorithms by providing smoother control signals an‎d lower energy consumption. This finding confirms the high potential of neuromorphic approaches in complex robotic control applications.

5