Residual Policy Learning

Ask about this topic

Overview

Residual policy learning studies how to improve an existing controller or learned policy by adding a smaller corrective policy, rather than replacing the whole system. Common approaches use reinforcement learning, imitation or human feedback to train bounded action corrections, often on top of frozen vision-language-action policies, classical controllers, or simulation-trained skills. These methods are applied to settings such as robot locomotion, contact-rich manipulation, and adaptation to faults or changing dynamics, with evaluation spanning simulation benchmarks and real-robot trials. A central challenge is giving corrections enough authority to handle unfamiliar conditions while preserving the base policy’s competence, safety, and stability.

Latest papers

CardsList

67 papers11 papers in the last four weeks, down 21% on the four weeks before. 0.1% of all new papers.

  1. From Prior to Pro: Efficient Skill Mastery via Distribution Contractive RL Finetuning

    Mar 10, 2026Zhanyi Sun, Shuran SongRL for RoboticsRobot Policy Learning

    1Stanford University.

  2. Stability-Aligned Residual Adaptation for Rapid Recovery from Dynamics Shifts

    Mar 8, 2026Nethmi Jayasinghe, Diana Gontero, Francesco Migliarba +4Robot Policy AdaptationResidual Policy Learning

    Department of Electrical and Computer Engineering, University of Illinois Chicago, Chicago, IL, USA. · Department of Neurobiology, Northwestern University, Evanston, IL, USA. · Department of Materials Science and Engineering, Northwestern University, Evanston, IL, USA. +2

  3. Residual RL-MPC for Robust Microrobotic Cell Pushing Under Time-Varying Flow

    Mar 5, 2026Yanda Yang, Sambeeta DasRobot Trajectory TrackingContact-Rich Robotic Manipulation

    Department of Mechanical Engineering, University of Delaware, Newark, DE 19716, USA

  4. Force-Aware Residual DAgger via Trajectory Editing for Precision Insertion with Impedance Control

    Mar 4, 2026Yiou Huang, Ning Ma, Weichu Zhao +4Interactive Imitation LearningContact-Rich Robotic Manipulation

    All authors are with Xi’an Jiaotong-Liverpool University, Suzhou, China.

  5. SPARR: Simulation-based Policies with Asymmetric Real-world Residuals for Assembly

    Feb 26, 2026Yijie Guo, Iretiayo Akinola, Lars Johannsmeier +3Robotic AssemblySim-to-Real Transfer

    Nvidia · University of Washington

  6. Physics-Guided Residual Reinforcement Learning for Humanoid Narrow-Path Traversal

    Aug 28, 2025Tianchen Huang, Sisheng Chen, Wei Zhou +8Humanoid Robot LocomotionRobot Skill Learning

    Institute of Humanoid Robots, Department of Precision Machinery and Precision Instrumentation, University of Science and Technology of China, Hefei, Anhui 230026, China

  7. PA-LOCO: Learning Perturbation-Adaptive Locomotion for Quadruped Robots

    Jul 5, 2024Zhiyuan Xiao, Xinyu Zhang, Xiang Zhou +1RL for Legged LocomotionQuadruped Locomotion

    School of Aeronautics and Astronautics, Sun Yat-sen University, Shenzhen 518107, P.R. China.