Residual Deep Reinforcement Learning-Based Computed Torque Control for a Cable-Driven Lower-Limb Rehabilitation Robot under Disturbances and Parametric Uncertainties

Abstract

This manuscript presents a simulation-based control framework for a planar three-DOF cable-driven lower-limb rehabilitation robot. A computed torque controller provides the model-based baseline, while a bounded residual DDPG policy compensates for tracking errors caused by disturbances and parametric uncertainties. Representative combined-case results show an approximately 42 percent reduction in RMS Cartesian tracking error compared with the baseline controller.

Citation

@article{fakouri2026rehabrobot,
  title = {Residual Deep Reinforcement Learning-Based Computed Torque Control for a Cable-Driven Lower-Limb Rehabilitation Robot under Disturbances and Parametric Uncertainties},
  author = {Fakouri, Mohammad-Hossein and Keymasi-Khalaji, Ali},
  journal = {arXiv preprint; under review at Expert Systems with Applications},
  year = {2026},
  arxiv = {2608.26739}
}