Building robots that can truly interact with the physical world.
Researcher at BAAI working on embodied intelligence. Our long-term vision is to build robots that can truly interact with the physical world — a vision we refer to as AETHER: Agile Embodied inTelligence for Humanoid-Environment InteRaction.
We are hiring both on-site and remote research interns, as well as full-time researchers and engineers, across embodied AI topics ranging from WBC to VLA. If you are highly self-motivated and deeply passionate about robotics, please feel free to contact me.
Pre-release preview · 2026 · arXiv coming soon
THOR2 is an upcoming general motion tracking model for humanoid robots, designed to help humanoids execute richer, more responsive motions for interactive, contact-rich tasks.
arXiv · 2026 · Corresponding author
BifrostUMI connects natural robot-free human demonstrations to executable humanoid whole-body skills through sparse keypoint learning, spatial retargeting, and whole-body control.
Nature Communications · 2025 · 16(1): 7893 · SCI (IF=15.7, CAS Q1, Top)
Contribution: data acquisition system, compliant control, full-system integration & debug, and model exploration.
IEEE Transactions on Human-Machine Systems · 2025 · 55(2): 165–175 · SCI (IF=4.4, CAS Q2, JCR Q1) · First author
Contribution: problem formulation, method design, experiments, analysis, writing & submission.
IEEE/ASME Transactions on Mechatronics · 2022 · 27(6): 4462–4474 · SCI (IF=7.3, CAS Q1, Top journal) · First author
Contribution: scientific question, method, experiments, analysis, writing & submission.
Selected systems I built, integrated, tuned, and deployed beyond publication-facing research.
CCTV-2 coverage from the 2024 World Robot Conference, featuring the robotic ultrasound system.
Additional lab demonstration of the autonomous ultrasound scanning platform.
Debugging vlog from Franka-based ultrasound robot integration, including real-robot testing and deployment refinement.
Embodied AI for Medical Robotics
This was my first project after completing my PhD. Working with colleagues, we built and integrated the engineering backbone that moved the system from lab prototypes toward real-world medical robotics demonstrations. My main contributions included:
Built data collection workflows for ultrasound probe guidance and human-robot interaction data.
Developed safe, intention-aware interaction and compliant control for contact-rich ultrasound scanning.
Handled real-robot debugging, structural design, cross-arm deployment, and touchscreen interaction UI.
Contributed to parts of model training, deployment, evaluation, and the end-to-end experimental pipeline.
Hydraulic manipulator platform developed during my PhD, later extended toward a dual-arm heavy-duty hydraulic robot.
Hands-on debugging vlog from the hydraulic manipulator build, including low-level servo tuning and real-robot integration.
Hydraulic Manipulation and Shared Control
Since 2018, I worked on a hydraulic manipulator platform from its early structural design stage through electrical control, low-level servo tuning, kinematics, teleoperation data collection, and shared-control research. The platform later evolved into a dual-arm heavy-duty hydraulic robot deployed at the State Key Laboratory of Fluid Power and Mechatronic Systems, Zhejiang University.
Independently designed, selected, integrated, and debugged the electrical control stack for hydraulic actuation.
Tuned low-level PID loops and controller parameters across valves, sensors, and hydraulic modules.
Implemented forward/inverse kinematics and connected the teleoperation data acquisition pipeline.
Proposed shared control to improve teleoperation efficiency and safety while reducing operator cognitive load.
Unlike modern electric manipulators, which often rely on highly integrated joint modules with built-in sensing, power electronics, communication, and low-level control interfaces, hydraulic manipulators require system-level integration across the entire actuation chain. Sensors, servo valves, hydraulic power supplies, electronic controllers, and servo loops must be co-designed and carefully tuned, making the final robot behavior a result of the coupled dynamics of the whole hydraulic system rather than an encapsulated joint module.
Undergraduate Freescale smart car competition project, awarded First Prize in the Western China regional contest.
Undergraduate Embedded Robotics
Around 2015, I participated in the Freescale smart car competition during my undergraduate study and won First Prize in the Western China regional contest. My main responsibilities covered structural design, embedded electronics, and controller development, giving me an early full-stack engineering foundation before my later work on hydraulic and medical robotic systems.
Worked on the mechanical layout, chassis integration, sensor placement, and competition-oriented reliability.
Designed and fabricated the motor driver board and main control board myself in Altium Designer, then completed hardware integration and debugging.
Developed and tuned low-level control logic for stable, fast, and repeatable vehicle motion.
Iterated the full system under real contest constraints, balancing speed, robustness, and maintainability.
Full list: Google Scholar.
