Research Analyzer
← Back ICRA 2026

Reinforcement Learning-Based Robust Wall Climbing Locomotion Controller in Ferromagnetic Environment

Yong Um, Young-Ha Shin, Joon-Ha Kim, Soonpyo Kwon, Hae-Won Park

PDF

AI summary

Key figure (auto-extracted from paper)
A three-phase curriculum learning strategy combined with realistic magnetic adhesion modeling enables quadrupedal robots to robustly climb vertical steel surfaces despite intermittent foot detachment, outperforming traditional model-based controllers.
reinforcement learning magnetic climbing quadrupedal locomotion curriculum learning sim-to-real transfer adhesion uncertainty

Problem

Traditional magnetic climbing controllers rely on model predictive control that assumes perfect adhesion, causing failures when surface imperfections or partial contact reduce holding forces. Naive reinforcement learning struggles with sim-to-real gaps and catastrophic falls during vertical climbing training.

Approach

The authors train a reinforcement learning policy using a three-phase curriculum that gradually transitions from flat-ground crawling to vertical climbing while introducing stochastic magnetic adhesion failures, paired with a physics-based model of electropermanent magnet foot contact.

Key results

  • Multi-phase curriculum enabling stable ground-to-wall gait transition
  • Physics-based stochastic adhesion model capturing partial contact and air-gap sensitivity
  • High success rate and rapid slip recovery in simulation under degraded adhesion
  • Successful untethered hardware experiments demonstrating robust vertical crawling on steel

Why it matters

Enables reliable deployment of magnetic climbing robots for industrial inspection and maintenance on complex or imperfect steel infrastructure.

Abstract

We present a reinforcement learning framework for quadrupedal wall-climbing locomotion that explicitly ad- dresses uncertainty in magnetic foot adhesion. A physics-based adhesion model of a quadrupedal magnetic climbing robot is incorporated into simulation to capture partial contact, air-gap sensitivity, and probabilistic attachment failures. To stabilize learning and enable reliable transfer, we design a three-phase curriculum: (1) acquire a crawl gait on flat ground without adhesion, (2) gradually rotate the gravity vector to vertical while activating the adhesion model, and (3) inject stochastic adhesion failures to encourage slip recovery. The learned policy achieves a high success rate, strong adhesion retention, and rapid recovery from detachment in simulation under degraded adhesion. Compared with a model predictive control (MPC) baseline that assumes perfect adhesion, our controller maintains locomotion when attachment is intermittently lost. Hardware experiments with the untethered robot further confirm robust vertical crawling on steel surfaces, maintaining stability despite transient misalignment and incomplete attachment. These re- sults show that combining curriculum learning with realistic adhesion modeling provides a resilient sim-to-real framework for magnetic climbing robots in complex environments.

Index terms

Legged Robots Climbing Robots Reinforcement Learning

Related papers