Research Analyzer
← Back ICRA 2026

A Practical Multi-Body Model Enabling a Flexible-Wheeled Robot to Learn Blind Stair Climbing

Chan-Young Yoon, Baek-Kyu Cho

PDF

AI summary

Key figure (auto-extracted from paper)
A simplified multi-body model combined with reinforcement learning enables a flexible-wheeled robot to successfully climb stairs using only proprioceptive feedback.
Reinforcement learning flexible wheels multi-body modeling blind control sim-to-real transfer soft robotics

Problem

Controlling flexible-wheeled robots for complex tasks like stair climbing is hindered by the nonlinear dynamics of soft materials, creating a difficult trade-off between simulation accuracy and reinforcement learning speed. Achieving robust, sensor-free control without visual feedback remains particularly challenging.

Approach

The wheel's continuous deformation is approximated as a Mass-Spring-Damper system of rigid links and joints, enabling fast simulation in NVIDIA Isaac Gym. A reinforcement learning policy is trained using only proprioceptive feedback, with dynamics randomization applied to bridge the sim-to-real gap.

Key results

  • Novel Mass-Spring-Damper multi-body model for efficient flexible wheel simulation
  • First end-to-end RL-based blind stair-climbing policy for a flexible-wheeled robot
  • Successful sim-to-real transfer enabling an 18 cm step climb (51% of wheel radius)
  • Robust policy generalization validated via dynamics randomization and cross-simulator testing

Why it matters

Provides a practical, computationally efficient framework for training robust locomotion policies in soft and hybrid robotics, accelerating real-world deployment in unstructured environments.

Abstract

Controlling a flexible wheeled robot for complex tasks such as stair climbing is highly challenging. The nonlinearity inherent in soft materials hinders accurate modeling, creating a trade-off in Reinforcement Learning (RL) between simulation fidelity and learning speed. We propose an RL-friendly, multi-body model that approximates the deformation of the flexible wheel as a Mass-Spring-Damper (MSD) system composed of rigid links and joints. This model enables end-to-end RL within a fast rigid-body simulator, facilitating a blind control policy that relies solely on pro- prioceptive feedback. To reduce the reality gap and enhance policy robustness, we randomize the main parameters of the MSD system. In real-world experiments, a robot successfully climbed an 18 cm step, corresponding to approximately 51% of the wheel radius—a feat impossible for a rigid-wheeled equivalent. To our knowledge, this is the first successful application of RL-based blind control for stair climbing with a flexible wheeled robot. However, structural limitations in our model and challenges in parameter identification hinder sim-to-real transfer, and improving robustness remains a key issue for future work.

Index terms

Reinforcement Learning Wheeled Robots Flexible Robotics

Related papers