IronMind: Scaling Humanoid Dexterous Manipulation via Camera-Space Ego-Centric Pretraining
arXiv preprint, September 2026
About
I am an M.S. Robotics student working on robot learning, world models, reinforcement learning, and scalable data systems for embodied intelligence. I am especially interested in closing the gap between learning algorithms and reliable deployment on real robots.
University of Michigan · Previously at XPENG Robotics, Lightwheel, and OpenRobotLab
Publications
Principal Steering Subspaces for Online Adaptation of Frozen Generative Robot Policies
arXiv preprint, September 2026
NeuralMetric: An accurate and efficient real-time safety metric for automated driving systems
Transportation Research Part A: Policy and Practice, vol. 210, article 105019, August 2026
Selected Projects
PiperPilot
An end-to-end, low-latency XR teleoperation and LeRobot-compatible data-collection toolkit for Piper robots, with Quest 3 and SpaceMouse control, replay, and policy deployment.
InternManip
A unified training and evaluation framework spanning ACT, Diffusion Policy, Pi0, and GR00T-N1/N1.5 across GenManip-v1, CALVIN, and SimplerEnv.
Large Reward Models for Robotic Manipulation
A trajectory-level reward and value model trained from 100K+ human-teleoperation episodes, enabling failure detection, data quality control, and improved VLA optimization.
Experience
World Models & Real-World RL
Developed learning and systems infrastructure for embodied robotic manipulation, including stereo representation learning, low-latency closed-loop action execution, and safety support for long-running hardware experiments.
Reward Modeling for Manipulation
Built large-scale teleoperation data and reward-modeling pipelines, improving VLA task success by about 30% over behavior cloning.
Unified Robot Learning
Co-developed InternManip and co-organized the IROS 2025 Multimodal Robot Learning Challenge in simulation and the real world.
Autonomous-Driving Safety
Developed NeuralMetric, published in Transportation Research Part A in 2026.
Education
University of Michigan
M.S. in Robotics · Expected Dec 2026
B.S. in Computer Science and Engineering · 2025
GPA 3.90/4.00 · Summa Cum Laude · Jackson and Muriel Lum Undergraduate Scholarship
Shanghai Jiao Tong University
B.S. in Electrical and Computer Engineering · 2025
GPA 3.81/4.00 · Distinguished Graduate · Undergraduate Outstanding Scholarship
Contact
I am happy to connect about robot learning, world models, real-world reinforcement learning, and research collaborations.