Robot policy improves precision in connector insertion tasks
ContactDP: Contact-Guided Diffusion Policy for Tight Insertion Tasks
Robotics
Summary
Inserting tiny parts together precisely is hard for robots because they can't fully see or feel the parts during the process. The authors created ContactDP, a smart robot policy that uses camera images, touch sensors on fingers, and force sensors to better understand how parts contact each other. This helps the robot adjust its movements smoothly to fit parts together even when things aren't perfectly lined up. Their method works better and is more reliable than systems that just use vision.
What this means in practice
- •For industrial automation engineers: Enable robots to perform precise electrical connector assembly with better correction from multiple sensors under uncertainty.
- •For manufacturing robotics teams: Improve reliability of robot insertion tasks for complex parts with varying shapes and misalignments by integrating tactile and force feedback.
Authors
Chengyi Xing, Shaoxiong Yao, Diego Romeres, Devesh K. Jha
Abstract
High-precision connector insertion remains challenging for robotic systems due to tight mechanical tolerances, partial observability during contact, and multimodal uncertainty arising from occlusion and contact ambiguity. Successful insertion requires closed-loop contact guidance that continuously integrates global alignment cues with local contact feedback to produce stable corrective actions under interaction. In this work, we present ContactDP (Contact-Guided Diffusion Policy for Tight Insertion Tasks), a multimodal diffusion-policy framework for contact-rich insertion. ContactDP jointly integrates wrist RGB observations, fingertip tactile sensing, and wrist-mounted force-torque measurements to infer contact state and generate temporally consistent corrective motions during insertion. To ensure stable execution under contact, the learned policy operates together with a hybrid position-force controller that provides compliant low-level interaction. We evaluate our approach on a suite of industrial-grade connector insertion tasks with varying connector geometries, grasp conditions, and initial misalignment. Across all tasks, ContactDP significantly outperforms vision-only diffusion policies for performance, reliability and generalization.