Researchers use deep RL to stabilize teleoperation robots with flexible joints

Teleoperation systems in critical domains like surgery, space exploration, and rehabilitation must handle joint flexibility and unpredictable communication delays. These two factors together often destabilize standard controllers, leading to unsafe oscillations and loss of coordination between master and slave robots. To address this, researchers from the University of Tehran (Armin Attarzadeh, Mohammad Ali Ghaemifar, Alireza Khanzadeh, Soheil Ganjefar) published a paper on arXiv (2607.21145) presenting a novel hybrid control architecture.

Their method pairs a classical Proportional-plus-Damping (P+d) controller with a deep reinforcement learning agent based on the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. The P+d controller provides guaranteed stability under bounded time-varying delays, while the TD3 agent continuously adapts the remote-side proportional and damping gains in real time. This reduces vibrations and improves trajectory tracking without requiring an explicit system model. The team proved stability using Lyapunov-Krasovskii analysis and shared source code on GitHub. The work offers a practical, adaptive solution for next-generation teleoperation systems.