Transferring reinforcement learning policies from simulation to Real-World robots remains a major challenge, particularly when dealing with low-level torque control, where even small modelling inaccuracies can lead to unstable or unsafe behaviours. In this work, we propose a Real2Sim2Real pipeline that improves Sim2Real transfer for torque-controlled robotic arms by combining trajectory matching, parameter optimization via genetic algorithms, and domain randomization. Using the 7-DOF Franka Emika Panda robot, we first identify friction, inertia, and gravity compensation parameters by minimizing the error between real and simulated joint trajectories. These calibrated dynamics are then used to train a TQC-based reinforcement learning agent in simulation. The trained policy is evaluated in both Gazebo and MuJoCo environments, and finally deployed on the real robot. Our results demonstrate a significant improvement in tracking accuracy and policy robustness after parameter tuning, with smooth policy transfer from simulation to the Real-World across multiple target-reaching tasks. This work highlights the effectiveness of accurate physical modelling in enabling stable and generalizable torque-based reinforcement learning policies.
Davide Bargellini, Alex Pasquali, Andrea Govoni et al.· 2026 IEEE/ASME International...· 0 citations
The automation of Deformable Linear Object (DLO) manipulation remains a key challenge in industrial production, where tasks such as connector assembly are still largely performed manually. Although prior work has demonstrated reliable wire terminal pose estimation and insertion monitoring using vision and tactile sensing, these approaches typically assume a fixed and known connector pose, limiting their applicability in flexible manufacturing scenarios. This paper presents a dual-arm robotic system for fully autonomous connector assembly. Two UR5e manipulators are employed: one dedicated to wire perception and insertion, and the other to connector localization and manipulation. To address the limitations of vision-only pose estimation, a multi-stage connector pose estimation strategy is proposed. First, a coarse 6D pose is obtained using a deep learning model trained on synthetic CAD data. Subsequently, the pose is refined through force-controlled contact scanning leveraging a wrist-mounted six-axis force/torque sensor. The acquired contact points are registered to the nominal CAD model via Iterative Closest Point, resulting in sub-millimeter localization accuracy. This connector pose strategy is seamlessly integrated with stereo-based wire perception and tactile-guided insertion monitoring, including pull-test validation. Experimental results demonstrate that the proposed system effectively compensates for connector pose uncertainty, achieving a 93.3% connector-level success rate and confirming its suitability for autonomous wire harness manufacturing.
M. Mirto, Alessio Caporali, Ž. Gosar et al.· 2026 IEEE/ASME International...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
