Research Analyzer
← Back ICRA 2026

DeepSkate: Reinforcement Learning of a Robust Controller for Energy Efficient Quadruped Skating

James Florin Petri, Gerard Lacey

PDF

AI summary

Key figure (auto-extracted from paper)
A model-free reinforcement learning controller enables a quadruped with passive wheels to skate efficiently and robustly across diverse terrains, cutting energy consumption by over 40% compared to existing methods.
Reinforcement Learning Quadruped Robots Passive Wheeled Locomotion Cost of Transport Sim-to-Real Transfer Legged Robotics

Problem

Standard legged robots suffer from poor energy efficiency, while passive wheeled-legged skating systems offer great potential but remain difficult to control due to complex dynamics, lack of active steering or braking, and terrain adaptability challenges.

Approach

The authors train an end-to-end reinforcement learning policy in simulation with custom rewards and domain randomization, then transfer it to a physical quadruped equipped with non-steerable passive wheels to learn agile, terrain-adaptive skating gaits.

Key results

  • Achieved stable skating speeds up to 1.5 m/s with 95% velocity tracking accuracy
  • Reduced cost of transport by 40.9% compared to skating state-of-the-art and 70.9% versus standard legged locomotion
  • Successfully navigated flat, rough, sloped, and stepped terrains without active braking or steerable ankles
  • Established a new performance benchmark for efficient quadruped skating locomotion

Why it matters

Proves passive wheeled-legged skating is a highly efficient and practical locomotion mode for quadrupedal robots, paving the way for longer-range urban and mixed-terrain deployments.

Abstract

Wheeled-legged hybrid robots have generated grow- ing interest in the research community due to the need for more efficient and versatile locomotion. Most recent research has focused on active wheels, but passive wheeled systems have great potential in improving energy efficiency. However, skating remains highly complex due to the difficulties of balancing dynamic motion, managing wheel-ground interactions, achieving precise torque control for smooth rolling, and adapting to unpre- dictable terrain while maintaining stability. We present an end- to-end model-free reinforcement learning approach that enables quadrupedal robots to skate efficiently, achieving agile and robust locomotion on both flat and rough terrain. Our skating- specific policy and sim-to-real pipeline are validated on a physical quadruped across diverse terrains with varying roughness, slopes, and features, consistently demonstrating controlled and efficient traversal. The robot achieves velocities up to 1.5 m/s with a cost of transport 40.9% lower than the skating state of the art and 70.9% lower than standard legged locomotion. These results establish skating as a feasible and efficient alternative mode of urban locomotion for quadrupedal robots, setting a foundation for future wheeled-legged research.

Index terms

Reinforcement Learning Legged Robots Sensorimotor Learning

Related papers