Research Analyzer
← Back ICRA 2026

Learning to Drift with Individual Wheel Drive: Maneuvering Autonomous Vehicle at the Handling Limits

Yihan Zhou, Yiwen Lu, Bo Yang, Jiayun Li, Yilin Mo

PDF

AI summary

Key figure (auto-extracted from paper)
A reinforcement learning framework with GPU-accelerated simulation and domain randomization enables robust, sim-to-real autonomous drifting control on an open-source Individual Wheel Drive platform.
Autonomous drifting Reinforcement learning Individual wheel drive Sim-to-real transfer GPU-accelerated simulation Agile motion control

Problem

Autonomous drifting control struggles with a significant simulation-to-reality gap and high training costs, while existing research platforms lack the independent wheel actuation needed for advanced maneuvering at handling limits.

Approach

The authors train a reinforcement learning policy in a custom GPU-accelerated parallel simulator using systematic domain randomization, then deploy it on a newly designed open-source 1/10 scale Individual Wheel Drive RC car.

Key results

  • Zero-shot sim-to-real transfer of drifting policies without real-world fine-tuning
  • Open-sourced 1/10 scale Individual Wheel Drive RC car platform with independent torque vectoring
  • GPU-accelerated simulator reducing reinforcement learning training time from hours to minutes
  • Precise trajectory tracking and stable sideslip control across complex maneuvers like direction reversals and variable-curvature paths

Why it matters

Advances safe autonomous vehicle maneuvering at friction limits and provides an accessible hardware-software benchmark for agile control research.

Abstract

Drifting, characterized by controlled vehicle motion at high sideslip angles, is crucial for safely handling emergency scenarios at the friction limits. While recent reinforcement learning approaches show promise for drifting control, they struggle with the significant simulation-to-reality gap, as policies that perform well in simulation often fail when transferred to physical systems. In this letter, we present a reinforcement learning framework with GPU-accelerated parallel simulation and systematic domain ran- domization that effectively bridges the gap. The proposed approach is validated on both simulation and a custom-designed and open- sourced 1/10 scale Individual Wheel Drive (IWD) RC car platform featuring independent wheel speed control. Experiments across various scenarios from steady-state circular drifting to direction transitions and variable-curvature path following demonstrate that our approach achieves precise trajectory tracking while maintain- ing controlled sideslip angles throughout complex maneuvers in both simulated and real-world environments.

Index terms

Motion Control Reinforcement Learning

Related papers