A Reinforcement Learning Based FEM Solver for Accelerating Contact-Influenced Simulation of Continuum Robots
Hao Chen, Jian Chen, Zhongkai Zhang, Hongbin Liu
Abstract
Continuum robots exhibit exceptional flexibility and multi-degree-of-freedom maneuverability, offering significant advantages for navigating confined luminal spaces. However, rapid simulation of their contact-influenced behavior remains challenging. This paper presents RLFEM, an innovative finite element method (FEM) featuring a reinforcement learning-enhanced quadratic programming solver, specifically designed for efficient continuum robot simulation updates. To address complex robot dynamics, we employ a quasi-static FEM formulation. Our core contribution integrates an accelerated solver scheme within this framework, leveraging reinforcement learning to rapidly resolve FEM contact problems without accuracy compromise. RLFEM exhibits significant gains in runtime performance for continuum robot simulations, as evidenced by numerical experiments showing a 16.20-fold acceleration compared to the baseline at the 150-node discretization level.
BibTeX
@inproceedings{ral2026_areinforcementle,
title = {A Reinforcement Learning Based FEM Solver for Accelerating Contact-Influenced Simulation of Continuum Robots},
author = {Hao Chen and Jian Chen and Zhongkai Zhang and Hongbin Liu},
booktitle = {RA-L 2026},
year = {2026}
}