CV
Education
Tsinghua University, incoming direct PhD student in Engineering Physics, starting 2026.09
Tsinghua University Weiyang College, B.S. in Mathematical Sciences and B.Eng. in Engineering Physics, 2022.08 - 2026.06
Research and Engineering Interests
- Embodied AI and vision-language-action models
- Robot imitation learning and real-robot deployment
- Robot data collection, rollout evaluation, and failure analysis
- Scientific simulation and physical-system modeling
Selected Projects
SO-ARM101 VLA Real-Robot Deployment, 2026
- Built a teleoperation-to-deployment loop for real-robot pickup-and-putdown.
- Worked with LeRobot/OpenPI, ACT, Diffusion Policy, SmolVLA, and pi0/pi0.5 deployment validation.
- Focused on dataset schema, normalization, relative action, action chunk, latency, rollout evaluation, and failure taxonomy.
Dual-Arm Hazardous Material Placement Prototype, 2026
- Built prototype components around ROS2 nodes, upper-computer interface, LeRobot interface, data collection scripts, state-action schema, teaching flow, and model training.
- Boundary: prototype and training validation, not final industrial delivery.
RoboCup THMOS Humanoid Small Team, 2026
- Supported field deployment and debugging for humanoid robots.
- Worked on network/IP setup, remote-controller binding, startup flow, state-machine debugging, and field troubleshooting.
STM32 Smart Car, 2024
- Worked on embedded robot control with FSM and PID, including obstacle avoidance, path planning, pushing, and shooting.
Scientific Simulation and Multimodal Systems, 2024 - 2026
- Experience with SSMB coherence simulation, full-cavity-enhanced Thomson scattering X-ray source design, HoloLens2 multimodal fusion, and ROS2 quadruped robot integration.
Skills
- Embodied AI / VLA: LeRobot, OpenPI, ACT, Diffusion Policy, SmolVLA, pi0, pi0.5, Transformer, RT-1/RT-2, RDT.
- Robot Systems: SO-ARM101, ROS2, MoveIt2, Webots, MuJoCo, FSM, PID, forward/inverse kinematics, Jacobian, trajectory planning.
- Engineering: Python, PyTorch, TensorFlow, Linux, Git, Docker, Hugging Face, OpenCV, NumPy, Pandas.
- Evaluation: success rate, stage pass rate, failure taxonomy, latency profiling, sim-to-real gap.
Links
- GitHub: https://github.com/xiaoms22
- SO-ARM101: https://github.com/xiaoms22/soarm101
- Portfolio repository: https://github.com/xiaoms22/embodied-ai-portfolio