Robot learning improves dressing by inserting wearable items precisely
Geometry-Aware Reinforcement Learning for Close-Fitting Wearable Insertion
Robotics
Summary
Putting on clothes or accessories with robots is hard because they need to fit just right to avoid getting stuck or missing the right spot. The authors see this as a problem of fitting one shape into another, like inserting fingers into a bracelet opening, and they focus on the important spatial relationships instead of every detail of the soft garment. They created a system that teaches a robot step by step how to align and insert the item, using feedback about each finger’s position to avoid mistakes. Their approach worked well in tests with both simulated and real robotic arms dressing a mannequin hand, showing improved success.
What this means in practice
- •For robotics engineers: Build robots that can dress people or mannequins by learning precise insertion of clothing or bracelets around body parts.
- •For prosthetics developers: Develop fitting procedures for wearable prosthetics or devices that require careful insertion around body parts using learned geometry-aware approaches.
Authors
Takuma Tsukakoshi, Elle Miller, Tamon Miyake, Namiko Saito, Yushi Wang, Shigeki Sugano, Tetsuya Ogata, Sethu Vijayakumar
Abstract
Robot-assisted dressing is particularly challenging due to narrow openings and articulated body parts. Even small errors in the relative pose between the body part and garment opening can cause snagging, missed insertion, or incomplete dressing. To address these challenges, we formulate dressing as a geometry-aware insertion problem that represents task-relevant spatial relationships between the body part and object opening, rather than relying on the full deformable-object state. Based on this formulation, we propose a Reinforcement Learning (RL) framework combining a staged curriculum, which progressively guides the policy from coarse alignment to insertion and final placement, with a finger-wise insertion soft gate that provides continuous feedback on individual finger insertion progress and helps suppress premature insertion. We evaluate the framework on rigid and deformable bracelet dressing tasks through component ablations, systematic variations in the initial hand pose, and closed-loop execution of the learned policies on a real robotic system. On the post-training deformable evaluation, the full method reaches 84.7 +- 1.2% success. The insertion gate accounts for the largest gain, and adding the curriculum on top of that gate reaches the top score. On a mannequin hand the same full policy succeeds on 15/20 real-robot trials.