CV
Education, research experience and skills.
Contact Information
| Name | Xinyao Li |
| Professional Title | Incoming MSR student, CMU Robotics Institute |
| xinyaoli511@gmail.com |
Professional Summary
Incoming MSR student at the Carnegie Mellon Robotics Institute, advised by Prof. Guanya Shi. BS in Computer Science from Shanghai Jiao Tong University (ACM Honors Class). Working on humanoid whole-body control and contact-rich object interaction.
Experience
-
2025 - 2026 Illinois, USA
Research Intern — ContactMimic
RoboVision @ Illinois Lab, University of Illinois Urbana-Champaign
Humanoid object interaction via contact control. Advised by Prof. Saurabh Gupta.
- Proposed a contact-conditioned keypoint-tracking policy that lets a humanoid make or suppress physical contact at test time by toggling per-body-part binary contact labels under the same reference motion.
- Designed two contact-following rewards and a trajectory-augmentation scheme to break the spurious correlation between keypoints and contact.
- Enabled task-specific object interaction (wiping, sitting, leaning on a backrest, box lifting) without any task-specific rewards.
- Trained in Isaac Lab and deployed on a real Unitree G1 humanoid.
- Primary contributor; submitted to CoRL 2026.
-
2025 - 2025 Beijing, China
Research Intern — efficient VLA
Robotics Group, ByteDance Research
Building a cost-effective and efficient generalizable VLA for robots. Supervised by Minghuan Liu.
- Achieved comparable performance with a 16x smaller model by distilling a pretrained VLA backbone, enabling shorter inference time and lower cost for deployment.
- Distilled the backbone by matching the hidden state latent space in Transformer layers, then finetuned with the action head.
- Validated with the CALVIN benchmark, OXE dataset (for pretraining) and the SIMPLER benchmark.
- Implemented the training and validation infrastructure as the independent developer.
-
2024 - 2025 Shanghai, China
Research Intern — RHINO
APEX Lab, Shanghai Jiao Tong University
Learning real-time humanoid-human-object interaction from human demonstrations. Advised by Prof. Weinan Zhang.
- Proposed the first real-time humanoid interaction framework capable of learning from human demonstrations, enabling dynamic task-switching and immediate responses to human instructions.
- Trained models directly from human demonstration videos via human pose estimation and retargeting.
- Developed a fully open-sourced 3D-printed motion capture system for easy reproduction of the pipeline.
- Developed the complete training pipeline and real-world deployment as a primary contributor.
- Second author with equal contribution. Accepted to IROS 2026.
Education
-
2026 - present Pittsburgh, PA, USA
MSR (Master of Science in Robotics)
Carnegie Mellon University
Robotics
- Advised by Prof. Guanya Shi, Robotics Institute.
-
2022 - 2026 Shanghai, China
BS, ACM Honors Class
Shanghai Jiao Tong University
Computer Science
- Computer Vision (100), Linear Algebra (97), Principle and Practice of Computer Algorithms (97), Computer Architecture (95), Advanced Compiler Designing (95), Machine Learning (93), Reinforcement Learning (93), Image Generation (93)
- A selective program admitting 30 students per year.
Skills
Robotics: Isaac Lab, MuJoCo, IsaacGym, 3D-printing, real-world deployment
Control: MPC, robot kinematics, adaptive control
Coding: Python, PyTorch, ROS2, C++