Research Analyzer
← Back ICRA 2026

RL-Based Coverage Path Planning for Deformable Objects on 3D Surfaces

Yuhang Zhang, Jinming Ma, Feng Wu

PDF

AI summary

Key figure (auto-extracted from paper)
A reinforcement learning agent using harmonic UV mapping and 2D feature extraction efficiently plans optimal wiping paths for deformable objects on complex 3D surfaces, outperforming traditional baselines.
Coverage Path Planning Reinforcement Learning Deformable Objects Harmonic UV Mapping Surface Wiping Robotic Manipulation

Problem

Traditional coverage path planning relies on rigid-environment assumptions and fails to handle the dynamic contact and deformation required when wiping or covering complex 3D surfaces with soft materials.

Approach

The method projects 3D surfaces into a simplified 2D space via harmonic UV mapping, processes real-time contact feedback with scaled grouped convolutions, and trains a reinforcement learning agent to output optimized coverage paths in simulation before real-world deployment.

Key results

  • Harmonic UV mapping reduces 3D state/action space for faster RL convergence
  • SGCNN extracts multi-scale contact features from 2D feature maps
  • Outperforms SPONGE and geometric baselines in path length and coverage area
  • Validated on Kinova Gen3 manipulator for real-world torso wiping

Why it matters

Enables reliable, contact-rich robotic manipulation of deformable objects for practical applications like medical rehabilitation, automated cleaning, and industrial surface treatment.

Abstract

Currently, manipulation tasks for deformable ob- jects often focus on activities like folding clothes, handling ropes, and manipulating bags. However, research on contact-rich tasks involving deformable objects remains relatively underdevel- oped. When humans use cloth or sponges to wipe surfaces, they rely on both vision and tactile feedback. Yet, current algorithms still face challenges with issues like occlusion, while research on tactile perception for manipulation is still evolving. Tasks such as covering surfaces with deformable objects demand not only perception but also precise robotic manipulation. To address this, we propose a method that leverages efficient and accessible simulators for task execution. Specifically, we train a reinforcement learning agent in a simulator to manipulate deformable objects for surface wiping tasks. We simplify the state representation of object surfaces using harmonic UV mapping, process contact feedback from the simulator on 2D feature maps, and use scaled grouped convolutions (SGCNN) to extract features efficiently. The agent then outputs actions in a reduced-dimensional action space to generate coverage paths. Experiments demonstrate that our method outperforms previous approaches in key metrics, including total path length and coverage area. We deploy these paths on a Kinova Gen3 manipulator to perform wiping experiments on the back of a torso model, validating the feasibility of our approach.

Index terms

Reinforcement Learning Motion and Path Planning

Related papers