6 August 2025 in IEEE Robotics and Automation Letters
Force-sensitive manipulation is essential for tasks such as cleaning, polishing, and surgical assistance, yet it remains challenging due to complex contact dynamics and the need for real-time adaptation. We propose DP-RRL, a hybrid learning framework that combines a diffusion policy (DP) for imitation learning (IL) with a residual reinforcement learning (RL) module to tackle these challenges. First, the DP leverages expert demonstrations to generate smooth, expert-like motion-force trajectories. Next, a residual RL component refines these trajectories online by adapting force commands to account for unmodeled contact dynamics. Our approach incorporates multimodal perception, including RGB point clouds, force feedback, and proprioceptive data, and achieves robust force-sensitive manipulation in unstructured environments. We validate DP-RRL on a 7-DOF robotic arm performing basin-cleaning tasks, where it demonstrates an average success rate of 88% across 12 unseen scenarios, outperforming state-of-the-art baselines.
1 October 2026 in IEEE Transactions on Vehicular Technology
Existing path planning methods often prioritize geometric shortest paths while lacking explicit control over safety margins, which may lead to unreliable navigation in complex environments. To address this issue, this paper proposes a Safety Margin Controllable Path Planning (SMCPP) framework. The key idea of SMCPP is to propagate safety constraints throughout the entire planning pipeline, ensuring that safety requirements are consistently maintained during search, path point filtering, and path smoothing. Unlike conventional planners that enforce safety through obstacle inflation after path generation, SMCPP integrates safety awareness directly into all planning stages and preserves it throughout the pipeline, rather than treating planning stages as independent modules. Extensive experiments demonstrate that SMCPP produces paths with larger obstacle clearance and smoother paths while achieving reliable navigation performance, particularly in narrow and cluttered environments. These results indicate that maintaining consistent safety constraints during planning significantly improves navigation robustness. Overall, SMCPP provides a practical and effective framework for safe and smooth autonomous navigation in complex environments.