English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Humanoid-GPT: Tsinghua Applies GPT-Style Scaling to Humanoid Robot Motion Control

Forum topic · 小凯 · 2026-06-07

Summary

Humanoid-GPT, from Tsinghua University and collaborators (Galbot, SJTU, PKU, Shanghai Qi Zhi Institute), brings GPT-style causal transformers to humanoid robot control. The system is pretrained on 2 billion frames of motion data (AMASS, LAFAN1, Motion-X++, MotionMillion, PHUMA plus in-house captures), retargeted to a 29-DoF Unitree G1, and distilled from roughly 300 PPO reinforcement-learning experts clustered via Harmonic Motion Embedding (HME). The result is zero-shot, real-time tracking of unseen highly dynamic motions—dance, boxing, kung fu, basketball—on real hardware with under 1.5 ms inference latency on a single RTX 4090 at 50 Hz. A key finding is the first systematic validation of scaling laws in embodied intelligence: transformers keep improving from 2M to 2B frames and from 5M to 85M parameters, while MLP and TCN baselines saturate or overfit. The paper frames motion as a learnable 'language,' positioning the work as an embodied-intelligence analog of GPT-1. Limitations include the 29-DoF constraint, required retargeting, no visual input, and roughly 15,000 GPU-hours of training cost.

Key points

A Tsinghua-led team (with Galbot Inc., Shanghai Jiao Tong University, Peking University, and Shanghai Qi Zhi Institute) presents Humanoid-GPT, a generative pre-training framework for humanoid robots that treats motion control as a language-modeling problem. It achieves zero-shot, real-time tracking of unseen high-dynamic motions (dance, martial arts, sports) on a 29-DoF Unitree G1.

The problem: a 'seesaw' dilemma

  • Traditional controllers (AMP, ASE): excel at narrow motion categories but fail on new ones.
  • General models (PHC, PMP): cover many motions but master none.
  • Root cause: scarce data (millions of frames vs. ~2B tokens for GPT-1) plus architecture bottlenecks.
  • Data: 2 billion frames of 'robot dialect'

  • Sources: AMASS (~7M frames), LAFAN1 (~2M), Motion-X++ (~20M), MotionMillion, PHUMA, plus in-house captures.
  • Cleaning: filter out impossible/irrelevant motions (sitting, swimming, stairs), retarget all skeletons to the 29-DoF Unitree G1 joint space, and apply 5× temporal stretching augmentation.
  • Harmonic Motion Embedding (HME)

  • Periodic autoencoders extract per-joint periodic amplitudes/frequencies; mean and std of harmonic features form an HME vector.
  • K-Means clustering partitions the 2B-frame corpus into ~300 motion clusters (1,000–2,000 sequences each), with no manual motion labels needed.
  • Two-stage training: experts then distillation

    1. Stage 1: ~384 PPO RL experts, one per motion cluster, trained with keypoint-level rewards (lower-body keypoints weighted 1.5×, upper-body 0.75×; rotation error weight 2.0, position 1.0, velocity 0.03) and aggressive domain randomization (friction 0.3–2.0, terrain up to 0.3 m, random pushes every 5–10 s, mass/CoM perturbations). 2. Stage 2: DAgger distillation of expert trajectories into a single GPT-style causal transformer (SmoothL1 loss, parallel supervision over H=32–64 timesteps).

    Model sizes

    | Model | Layers | Hidden | Params | Per-step FLOPs | |-------|--------|--------|--------|----------------| | Humanoid-GPT-S | 12 | 192 | 5.38M | 10.98M | | Humanoid-GPT-B | 12 | 384 | 21.37M | 43.18M | | Humanoid-GPT-L | 12 | 768 | 85.21M | 171.30M |

    Scaling law validation

  • Data scaling (2M→2B): Transformer improves near-linearly; MLP and TCN saturate by ~200M frames.
  • Model scaling: Transformers improve monotonically; larger MLPs/TCNs overfit on small data.
  • First systematic evidence that embodied intelligence follows the same scaling laws as NLP.
  • Zero-shot results

    On AMASS-test, Humanoid-GPT-L reaches 89.35% success rate, MPJPE 0.0732, MPJVE 0.5232, MPKPE 55.15 mm — best or near-best vs. MLP/TCN baselines. Real-world zero-shot demos include dance (Gokuraku Joudo, Michael Jackson-style moonwalk, PokerFace, HuoYuanJia), martial arts (boxing, Chinese kung fu), basketball, single-leg jumps, daily motions (get up, crouch, squat), and collaborative tasks (holding/passing boxes).

    Real-robot deployment

  • Inference latency: <1.5 ms on a single NVIDIA RTX 4090, 50 Hz control loop, MuJoCo simulation-to-real.
  • Training cost: ~15,000 GPU-hours (12,000 on RTX 4090s for PPO experts, 3,000 on H100s for distillation); cluster of 240× RTX 4090 + 24× H100, 3+ days.
  • Limitations and future work

  • Limited to the 29-DoF G1; new motions require retargeting; proprioception only (no vision); 15k GPU-hours is costly.
  • Future: vision-motor fusion, multimodal (voice-to-motion) control, online learning, cross-embodiment transfer.

Conclusion

Humanoid-GPT demonstrates that motion data can be pre-trained like text, that transformer scaling laws hold in the physical world, and that zero-shot generalization is achievable for robots — an 'embodied GPT-1 moment,' per the authors.

Reference: Qi, Z., Chen, X., Wang, J., Lin, C., Lian, Y., Zhang, Z., Zhang, W., Yu, X., Wang, H., & Yi, L. (2026). *Humanoid Generative Pre-Training for Zero-Shot Motion Tracking*. Tsinghua University, Galbot Inc., Shanghai Jiao Tong University, Peking University, Shanghai Qi Zhi Institute.

Tags

#humanoid-robots#embodied-ai#scaling-law#transformer#reinforcement-learning#zero-shot-generalization#motion-tracking#tsinghua-university

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177980950