UC San Diego
CoCo Matrix — Shanghai
Led the in-context learning runtime and on-robot deployment for WRC. Built unified multi-robot infrastructure for the GR3, dual-PiPER-X, and Astribot S1 platforms, wiring policy serving, perception, teleoperation, and safety into modular Dora dataflow graphs. Implemented latency-aware asynchronous inference with action-chunk scheduling that overlaps inference with execution while holding control-rate stability.
Hao Su's AI Lab — UC San Diego
Developing dexterous teleoperation frameworks, sim-to-real pipelines, and predictive world models.
MURO Lab — UC San Diego
Developing multi-robot map-merging infrastructure for autonomous TurtleBot4 fleets.
Triton AI — UC San Diego
LiDAR–camera fusion for FTENTH and path planning for an autonomous go-kart.
* Equal contribution·† Equal advising
Weikang Wan, Zeyu Lai, Jianyu Hou, Yuchen Zhou, Jiawei Fu, Henrik I Christensen†, Hao Su†
CoRL · 2026
A contact-aware foundation controller that turns upstream fingertip-motion drafts, from teleoperation or a learned policy, into contact-consistent joint commands. The same trained controller improves robustness without retraining, raising real-world teleoperation success from 29.2% to 78.3%.
Read PaperWeikang Wan, Jinqiu Wang, Jianyu Hou, Zeyu Lai, Henrik I Christensen†, Hao Su†
Preprint · 2026
A physics-grounded pipeline that converts a handful of human RGB-D videos into large-scale dexterous manipulation datasets under randomized layout, geometry, appearance, and physics. It improves simulated downstream policy success by 33.2% on average and reaches 45% zero-shot real-world success from only three human videos per task.
Read PaperApplied and Computational Engineering · 2024
This research explores the fundamental mechanics of virtual reality environments, focusing on computational engineering principles to enhance user immersion and system performance within virtual realms.
Read Paper