Humanoid Learning & Control
Translating language instructions into coordinated whole-body behaviors through dynamics-aware representations and robot control.
Haozhe Jia 贾浩哲
Incoming Ph.D. Student at Peking University EPIC Lab · Research Intern at Galbot
I am an incoming Ph.D. student at Peking University. I have joined EPIC Lab, where I work under the supervision of Prof. He Wang and Prof. Li Yi. I am currently a research intern at Galbot and an undergraduate at Shandong University.
My research focuses on humanoid robot learning and control, particularly on connecting generative motion models with physical execution. I am interested in how robots can translate language instructions into coordinated, physically grounded whole-body behaviors. My work spans motion generation, dynamics-aware representations, and language-conditioned humanoid control.
Research
Generative Motion Modeling
Learning language-conditioned motion representations and generative models for expressive and controllable movement.
Physics-Grounded Generative Models
Incorporating physical structure into generative modeling to improve consistency and generalization.
Selected Publications
Before the Body Moves: Learning Anticipatory Joint Intent for Language-Conditioned Humanoid Control
Introduces dynamics-aligned joint intent representations that anticipate support transfer, contact switching, and balance preparation for streaming language-conditioned humanoid control.
ECHO: Edge-Cloud Humanoid Orchestration for Language-to-Motion Control
Connects cloud-based motion generation with on-device closed-loop tracking for language-to-motion humanoid control, validated in simulation and on real hardware.
Learning to Think in Physics: Breaking Shortcut Learning in Scientific Diffusion via Representation Alignment
Proposes REPA-P to align denoising features with physics-aware representations, improving physical consistency and out-of-distribution robustness.
ANT: Adaptive Neural Temporal-Aware Text-to-Motion Model
Introduces step-aware temporal modulation and late-stage CFG reduction for diffusion motion models, improving semantic alignment and retrieval performance.
RMDM: Physics-Informed Representation Alignment for Sparse Radio-Map Reconstruction
Combines a PINN-based field initializer with a diffusion refiner to reconstruct sparse radio maps accurately under physically constrained settings.
More Publications
DCTdiff: Intriguing Properties of Image Generative Modeling in the DCT Space
Develops an end-to-end diffusion model in DCT space for higher-resolution image generation with stronger efficiency and spectral interpretability.
LUMA: Low-Dimension Unified Motion Alignment with Dual-Path Anchoring for Text-to-Motion Diffusion Model
Uses temporal semantic anchors and low-frequency motion anchors to improve deep U-Net alignment, gradient flow, and convergence in diffusion motion synthesis.
Towards Better Evaluation Metrics for Text-to-Motion Generation
Introduces OTMS and MMMD, two evaluation metrics designed to better correlate text-to-motion quality with human judgment.
POLARIS: Projection-Orthogonal Least Squares for Robust and Adaptive Inversion in Diffusion Models
Derives a theoretically grounded projection schedule for diffusion inversion, improving reconstruction quality without substantial extra computation.
RadioFlow: Efficient Radio Map Construction Framework with Flow Matching
Uses deterministic flow matching for fast radio map construction, reducing parameter count and inference time while maintaining reconstruction quality.
Free-T2M: Frequency Enhanced Text-to-Motion Diffusion Model With Consistency Loss
Introduces frequency-aware consistency supervision to stabilize motion denoising and improve semantic fidelity in diffusion-based text-to-motion generation.
Guided Path Sampling: Steering Diffusion Models Back on Track with Principled Path Guidance
Applies manifold-aware interpolation and dynamic guidance schedules to keep diffusion sampling closer to valid data trajectories.
Experience
Research Intern
Galbot
Embodied AI Algorithm Intern
LimX Dynamics
- Developed ECHO, a language-driven humanoid motion control system with a compact 38-DoF action representation.
- Built a cloud-edge streaming pipeline: cloud diffusion generates motion references; on-device lightweight controller performs closed-loop tracking.
- Validated in MuJoCo simulation and on real humanoid hardware.
Research Assistant
Hong Kong University of Science and Technology (Guangzhou)
- Led research on PhyRMDM, Free-T2M, and LUMA, spanning radio map reconstruction and text-driven human motion generation.
- Owned the full pipeline from model selection and training to ablation design; all code open-sourced.
- First-author / co-first-author publications at ICML and ACM MM (Oral).
Contact
Leave a Note
Feel free to leave a message or discuss collaboration ideas.
If comments do not load below, read or leave a note on GitHub.