Sim2real Transfer for Reinforcement Learning in Robotic Arm Control: a Closed-Loop Optimization approach for Parameter Estimation

This thesis delves into sim-to-real transfer challenges for torque-controlled robotic arms. Employing reinforcement learning, the research addresses the simulation-reality gap to improve control policies for practical applications. The study investigates static and dynamic friction compensation, as well as the optimization of simulation parameters and domain randomization. The results offer valuable insights into the efficacy of the proposed methodology and its potential implications for real-world robotic systems. Additionally, the thesis introduces an approach to refining simulation parameters by comparing trajectories generated in simulation with those obtained from real-world experiments. This contributes to a more comprehensive understanding of the sim-to-real transfer problem and advances the state-of-the-art in robotic control.