Advanced Input Shaping Control Based on Reinforcement Learning for SCARA Robot

In this study, a deep reinforcement learning-based input shaping technique (DRL-IST) is proposed to minimize residual vibration in industrial robot arms. The DRL-IST automatically tunes input shaper parameters to enhance vibration suppression following rapid point-to-point motions. The input-shaping design problem is formulated as a Markov Decision Process (MDP), and a deep deterministic policy gradient (DDPG) agent is employed to learn an optimal adaptation strategy for the shaper parameters. The proposed DRL-IST scheme is validated through simulations and experiments on a SCARA robot arm. Conventional two-impulse Zero-Vibration (ZV), three-impulse Zero-Vibration and Derivative (ZVD) shapers are implemented as benchmarks under various configurations. The results demonstrate that the DRL-IST significantly reduces residual vibrations and improves robustness against modeling uncertainties and external disturbances compared to conventional fixed-parameter input shaping, particularly in systems with configuration-dependent vibrations.