# HybridMimic: Hybrid RL-Centroidal Control for Humanoid Motion Mimicking

arXiv: https://arxiv.org/abs/2603.06775 · HTML: https://mlsanity.com/paper/2603.06775/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in RL for robot locomotion and manipulation, one of 15 directions in Deep reinforcement learning methods.
- Builds on 22 of its 40 references that are in the map.
- First posted on arXiv 6 Mar 2026.

## Placed in

- Deep reinforcement learning methods / [RL for robot locomotion and manipulation](https://mlsanity.com/deeprl/rl-for-robot-locomotion-and-manipulation/)

## Builds on (most cited first, 22 of 22)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World](https://mlsanity.com/paper/1703.06907/) (2017)
- [Sim-to-Real Transfer of Robotic Control with Dynamics Randomization](https://mlsanity.com/paper/1710.06537/) (2017)
- [Asymmetric Actor Critic for Image-Based Robot Learning](https://mlsanity.com/paper/1710.06542/) (2017)
- [Robust Motion In-betweening](https://mlsanity.com/paper/2102.04942/) (2021)
- [BeyondMimic: From Motion Tracking to Versatile Humanoid Control via Guided Diffusion](https://mlsanity.com/paper/2508.08241/) (2025)
- [Retargeting Matters: General Motion Retargeting for Humanoid Motion Tracking](https://mlsanity.com/paper/2510.02252/) (2025)
- [RL + Model-based Control: Using On-demand Optimal Control to Learn Versatile Legged Locomotion](https://mlsanity.com/paper/2305.17842/) (2023)
- [GLiDE: Generalizable Quadrupedal Locomotion in Diverse Environments with a Centroidal Model](https://mlsanity.com/paper/2104.09771/) (2021)
- [CAJun: Continuous Adaptive Jumping using a Learned Centroidal Controller](https://mlsanity.com/paper/2306.09557/) (2023)
- [SoftMimic: Learning Compliant Whole-body Control from Examples](https://mlsanity.com/paper/2510.17792/) (2025)
- [3D Underactuated Bipedal Walking via H-LIP based Gait Synthesis and Stepping Stabilization](https://mlsanity.com/paper/2101.09588/) (2021)
- [Template Model Inspired Task Space Learning for Robust Bipedal Locomotion](https://mlsanity.com/paper/2309.15442/) (2023)
- [RAMBO: RL-Augmented Model-Based Whole-Body Control for Loco-Manipulation](https://mlsanity.com/paper/2504.06662/) (2025)
- [Inverse Dynamics Control of Compliant Hybrid Zero Dynamic Walking](https://mlsanity.com/paper/2010.09047/) (2020)
- [Time-Varying ALIP Model and Robust Foot-Placement Control for Underactuated Bipedal Robot Walking on a Swaying Rigid Surface](https://mlsanity.com/paper/2210.13371/) (2022)
- [Enhancing Model-Based Step Adaptation for Push Recovery through Reinforcement Learning of Step Timing and Region](https://mlsanity.com/paper/2411.01000/) (2024)
- [Adaptive Ankle Torque Control for Bipedal Humanoid Walking on Surfaces with Unknown Horizontal and Vertical Motion](https://mlsanity.com/paper/2410.11799/) (2024)
- [Efficiently Learning Robust Torque-based Locomotion Through Reinforcement with Model-Based Supervision](https://mlsanity.com/paper/2601.16109/) (2026)
- [Data-driven Adaptation for Robust Bipedal Locomotion with Step-to-Step Dynamics](https://mlsanity.com/paper/2209.08458/) (2022)
- [Stair Climbing using the Angular Momentum Linear Inverted Pendulum Model and Model Predictive Control](https://mlsanity.com/paper/2307.02448/) (2023)
- [ExtremControl: Low-Latency Humanoid Teleoperation with Direct Extremity Control](https://mlsanity.com/paper/2602.11321/) (2026)

## Abstract (arXiv)

Motion mimicking, i.e., encouraging the control policy to mimic human motion, facilitates the learning of complex tasks via reinforcement learning (RL) for humanoid robots. Although standard RL frameworks demonstrate impressive locomotion agility, they often bypass explicit reasoning about robot dynamics during deployment, which is a design choice that can lead to physically infeasible commands when the robot encounters out-of-distribution environments. By integrating model-based principles, hybrid approaches can improve performance; however, existing methods typically rely on predefined contact timing, limiting their versatility. This paper introduces HybridMimic, a framework in which a learned policy dynamically modulates a centroidal-model-based controller by predicting continuous contact states and desired centroidal velocities. This architecture exploits the physical grounding of centroidal dynamics to generate feedforward torques that remain feasible even under domain shift. Using physics-informed rewards, the policy is trained to efficiently utilize the centroidal controller's optimization by outputting precise control targets and reference torques. Through hardware experiments on the Booster T1 humanoid, HybridMimic reduces the average base position tracking error by 13% compared to a state-of-the-art RL baseline, demonstrating the robustness of dynamics-aware deployment.