Reusing Locomotion Training Recipes Across Humanoids via Body-Aligned Virtual Joints
Authors: Sol Choi, Mingon Kim, Seokin Seo, Jongeun Choi, Donghyeon Kim, Yisoo Lee
Organizations: Center for Humanoid Research, Korea Institute of Science and Technology, 02792, Seoul, South Korea · Department of Mechanical Engineering, Yonsei University, 03722, Seoul, South Korea · Advanced Robotics Lab., CTO Division, LG Electronics Inc., 06772 Seoul, South Korea · 1X Technologies, 150 Industrial Rd, San Carlos, CA 94070, USA
Reinforcement learning has enabled robust humanoid locomotion, yet training recipes (rewards, architectures, hyperparameters) remain hard to reuse across robots with different native joint coordinates, since policies output native joint angles while action scales, bounds, gains, and penalties are defined in those robot-specific coordinates. Existing transmission models convert native targets into actuator commands but do not standardize the coordinate interface exposed to the policy. We introduce a body-aligned virtual joint space upstream of this mapping: policy actions and observations follow a common anatomical convention, and a deterministic kinematic layer maps virtual targets to native ones, leaving the existing transmission and low-level control stack unchanged. Under a single unmodified recipe on three humanoids with different leg-axis layouts, it is the only interface that tracks accurately, keeps the feet aligned with the travel direction, and trains stably on all three, and on Unitree G1 hardware it reduces heading drift by nearly a factor of six.
Figures & tables
Fig. 1 : Hip-joint coordinates of three representative humanoids. The native axes differ across platforms (top), whereas the proposed body-aligned virtual joint coordinates express the same joints about a shared body convention, giving a consistent axis layout across embodiments (bottom).
Fig. 2 : Overview of the proposed body-aligned virtual joint interface. The proposed modules map policy actions and joint-state observations between the shared virtual space and each robot’s native space, while the existing native-to-actuator control stack remains unchanged.
Fig. 3 : Humanoid platforms used in simulation—Unitree G1, PND Adam, and a custom-built platform—with their leg joint-axis configurations.
Robot
Interface
Velocity Error (m/s)
Lateral Drift (m)
Unitree G1
Native
0.042 ± 0.001
0.086 ± 0.001
Task space
0.048 ± 0.002
0.129 ± 0.003
Virtual (in-loop)
0.042 ± 0.007
0.084 ± 0.001
Virtual (post-hoc)
0.053 ± 0.018
0.089 ± 0.003
PND Adam
Native
0.106 ± 0.018
0.094 ± 0.007
Task space
0.207 ± 0.016
0.082 ± 0.010
TABLE I : Sim-to-sim velocity tracking on the constant velocity command ( n=3 seeds, mean ± std). Velocity error and lateral drift are expressed in the base heading frame.
Fig. 4 : Foot-progression angle (toe-out) by platform and coordinate interface (all seeds, both feet).
Fig. 5 : Swing-leg motion on the PND Adam in simulation under the native, proposed, and task-space interfaces. The proposed interface keeps the feet aligned with the direction of travel.
Fig. 6 : Training stability across the three embodiments, measured by mean episode length over the first 6 k iterations (mean ± standard deviation over three seeds). The dotted line marks half of the maximum episode length.
Fig. 7 : Hardware rollouts under a constant straight-walking command, composited from video frames of one representative rollout per interface. The dash-dotted line is a 10 m floor reference along the commanded heading; the solid curve traces the path walked.
Whole-body humanoid loco-manipulation requires coordinating the robot's entire kinematic chain. However, most existing systems typically decouple the upper and lower bodies into separate controllers, limiting such coordination and yielding behaviors similar to those of a wheeled dual-arm platform. In this paper, we ask what it takes to build a whole-body native vision-language-action (VLA) model that maps language and pixels directly to all of the humanoid's degrees of freedom. We conduct a systematic empirical study organized as a roadmap of one-variable-at-a-time experiments across three phases: whole-body teleoperation, VLA model design, and heterogeneous co-training. Our study yields several intriguing findings: a joint-based whole-body teleoperation interface outperforms alternatives that only partially expose the humanoid's degrees of freedom; a VLA pretrained on static and wheeled dual-arm platforms transfers surprisingly well to a humanoid's full action space; and co-training with HuMI, the humanoid analog of UMI, extends the policy to new objects and instructions without additional whole-body teleoperation on those targets. Following this roadmap yields OpenHLM, an open-source recipe for whole-body humanoid loco-manipulation. In a challenging long-horizon task that spans a wide vertical range of the humanoid, OpenHLM outperforms two state-of-the-art humanoid VLA baselines (GR00T N1.6 and Ψ0) using less than half the total demonstration time. Our code, training data, and model checkpoints are available at [https://openhlm-project.github.io/].
Yingdong Hu, Haodong Zhu, Boyuan Zheng +6
Tsinghua University · Shanghai Qi Zhi Institute · Spirit AI
Despite recent advances in humanoid locomotion, controllers optimized for command tracking and robustness tend to produce mechanical gaits, whereas controllers tied to human motion data often fail to generalize to commands outside the data distribution. This work introduces a learning framework that balances these competing objectives to synthesize real-time steerable, robust, and biomimetic locomotion policies from human data. Using an in-house curated locomotion dataset covering diverse speeds and directions, we first learn a natural locomotion prior policy through a teacher-student distillation process. Specifically, we train a full-body reference-conditioned policy with Reinforcement Learning (RL), then distill it into a lightweight prior policy conditioned solely on proprioception and a planar torso-velocity steering command. Next, we fine-tune the prior policy with multi-task RL to expand command coverage and robustness beyond the data distribution, pairing a goal-conditioned task that tracks arbitrary commands with a reference-guided task that tracks the human data as an explicit style regularizer. We validate our framework on three humanoid robots: the Boston Dynamics Atlas R1, Atlas D1, and Unitree G1. Experimental results demonstrate robust performance across real-world scenarios, including direct user-controlled locomotion in indoor and outdoor environments, and integration as the locomotion layer within hierarchical control stacks. Benchmarks against Tabula Rasa RL policies trained without human data and ablation studies confirm that our framework yields a lightweight, deployable policy that reconstructs coordinated whole-body behavior from a steering command, retaining the human gait characteristics while remaining robust and fully steerable.
Mike Zhang, Dongho Kang, Kevin Bergamin +19
RAI Institute, USA · Boston Dynamics, USA · Carnegie Mellon University, USA
Achieving human-like behavior on humanoid robots has advanced rapidly by leveraging motion-capture demonstrations rather than hand-crafted cost functions. Adversarial imitation supplies a style signal that lets a policy fulfill a task while adhering to a desired style, but is difficult to transfer to hardware. Whole-body tracking through reinforcement learning instead deploys reliably and imitates demonstrations faithfully, but the resulting policies are conditioned on the future reference, so they require a reference trajectory at runtime and can fail when the motion differs from the training data. We train a humanoid policy directly with whole-body tracking rewards while restricting its actor to compact locomotion commands and robot observations. Reference motion supervises the reward and privileged critic, without supplying reference joint trajectories to the actor or requiring teacher distillation. The demonstrated style is thereby encoded in the policy through the tracking reward, so the robot reproduces the whole-body style under commanded velocities. Trained on a single random-walk motion and commanded with velocity profiles from held-out motions, our policy reaches at least 98% success in simulation and completes every trial on the Unitree G1. On held-out slow walking, whole-body tracking drops to 2% success and command-conditioned distillation to 56%, and both fail on the real robot. On the trained motions, the joint position error of our policy stays within 8% of whole-body tracking, and it has a higher style coverage than adversarial motion priors (AMP), which fell on the real robot without a safety crane.
Murad Dawood, Oezge Drama Berkenkamp, Antonio Gonzales Marin +1
Agile Robots SE, Munich, Germany. · LS Wiiri Innovation Center, Wuxi, China.