ZiangZheng

About Me

I am an MPhil student at Tsinghua University (School of Vehicle and Mobility, iDlab), advised by Prof. Shengbo Eben Li. My research focuses on full-stack humanoid learning: building scalable reinforcement-learning systems for embodied agents, learning foundation behavior priors (adversarial / diffusion / generative), predictive world models, and closed-loop sim-to-real deployment on Unitree G1.

I am the founder of Renforce Dynamics (20+ open-source repos, 1,100+ stars) and the gait-group lead of the THMOS humanoid soccer team.

Research Interests

• Foundation Behavior Models for humanoid whole-body control: adversarial and diffusion motion priors, task-priority optimization, multimodal motion tracking.
• Predictive World Models for embodied agents: video-conditioned inverse dynamics, VLM-based reasoning under egocentric POMDP.
• Self-Improving RL: jump-start initialization, self-evolving motion priors, lifelong policy distillation.
• Cross-Embodiment Sim-to-Real: morphology-agnostic latent-to-latent policies, scalable training infrastructure (IsaacLab / IsaacSim / mjlab / Genesis).

📄 CV: latest (2026-05) · dated archive

News

• [2025-07] Released TrackerLab — an IsaacLab extension for multi-mode whole-body tracking.
• [2025-06] Learning a Shared Latent-to-Latent Locomotion Policy for Diverse Legged Robots accepted to IROS 2025 (First Author).
• [2025-04] Canonical Form of Datatic Description in Control Systems accepted to ACC 2025.
• [2025-03] Learn a Safe and Efficient MPC with Lagrange Multiplier Neural Network accepted to IEEE TCST.
• [2023] Adaptive Device Sampling and Deadline Determination accepted to Journal of Cloud Computing.

Selected Publications

Publications

Services

Open Source Community

• Renforce Dynamics - Leader

Powered by Jekyll and Minimal Light theme.