Learning Functional Compliance in Whole-body Manipulation
* Equal contribution.
I'm Zehua Wang (王泽华), an undergraduate at MIT pursuing a B.S. in Physics and Artificial Intelligence & Decision Making (GPA: 5.0/5.0; expected May 2028). My work sits at the intersection of physics and AI: from reinforcement learning for humanoid control to diffusion models and molecular dynamics.
At MIT, I co-developed Diffusion Accelerants with Prof. Tommi Jaakkola's group. At FortyFive Labs, advised by Dr. Ge Yang, I worked on reinforcement learning, functional compliance, and sim-to-real humanoid control on Unitree G1 hardware. In summer 2026, I joined Hudson River Trading for rotations in algorithm development and software engineering.
Previously, I worked with Prof. Huazhe Xu at Tsinghua University on learning-based robot manipulation and sim-to-real transfer. Before MIT, I completed preparatory and first-year coursework at Tsinghua's Institute for Interdisciplinary Information Sciences (IIIS). I won a gold medal and placed first in theory at the 54th International Physics Olympiad (IPhO 2024).
Reinforcement Learning Humanoid Whole-Body Control Sim-to-Real Transfer Imitation Learning Scaling Laws for Embodied AI Generative Models Diffusion Models
Outside of research, I'm into basketball (led the IIIS team as a freshman), skiing (10+ years), and lifting. I also enjoy soccer, volleyball, table tennis, swimming, and hiking. In my spare time, I listen to music and occasionally sing.
I'm always happy to chat, whether it's about research, physics, AI, or anything else. If something here caught your interest, feel free to reach out.
Published and presented research
* Equal contribution.
* Equal contribution.
Selected research projects and open-source work
I authored a three-commit patch that strengthens speech-to-text provider validation in NousResearch/Hermes Agent . The changes make voice-requirement checks recognize selected providers and plugin-backed speech-to-text implementations, with tests covering the new behavior. The contribution is now part of the repository's main branch with authorship preserved. The links above lead to the filtered main-branch history and the maintainer review that incorporated the patch.
6.8300 final project: uses a video model to hallucinate orbital observations from one populated-scene photo, then reconstructs the scene with a pose-free 3D stack.
Compared DDPM and Flow Matching for humanoid trajectory synthesis; at five function evaluations, Flow Matching achieved 820 vs. 280 survival steps.
Recent updates and announcements
Joined Hudson River Trading for algorithm development and software engineering rotations.
Started working with Prof. Tommi Jaakkola's group on diffusion models and molecular dynamics.
1st Place at MITIT 2025 Winter Contest (Beginner's Round).
Joined FortyFive Labs (Dr. Ge Yang) to work on reinforcement learning and embodied AI.
Started sophomore year at MIT.
Gold Medal at IPhO 2024 (1st place in the theoretical round).