CV

Bachelor of Engineering

Southern University of Science and Technology (SUSTech)

Electronic Information Engineering

• GPA: 3.90/4.00 · Rank: 3/46
• Selected Coursework: Data Structures and Algorithm Analysis, Probability and Statistics, Machine Learning, Robotic Motion and Control, Digital Signal Processing, Signals and Systems, Analog Circuits, Digital Circuits

Undergraduate Researcher

Robotics and Computer Vision Lab, SUSTech

Supervisor: Prof. Hong Zhang

• Topics: Social Navigation and Autonomous Driving
• Contributed to TPT-Bench data collection for social navigation; gained hands-on experience with ROS bag processing, multisensor synchronization, and data preprocessing (paper under review at IJRR)
• Independently proposed ProDrive, a proactive autonomous driving framework jointly training a driving planner and BEV world model via ego-environment co-evolution
• Demonstrated stronger performance than prior world-model-based driving frameworks on NAVSIM v1
• First author; led method design, experiments, and manuscript preparation. Accepted by CVPR 2026 GigaBrain Challenge Workshop

Undergraduate Researcher

HMI Lab, School of Computer Science, Peking University

Supervisor: Prof. Shanghang Zhang

• Topic: Vision-Language-Action Models, World Models, and Latent-Space Simulation
• Proposed Token-World, a latent world model for VLA systems predicting future visual tokens directly in token space, eliminating intermediate RGB reconstruction
• Designed a spatio-temporal multimodal Transformer with GRU-enhanced temporal modeling for action-conditioned long-horizon prediction
• Combined flow matching with diffusion forcing to improve rollout stability and reduce autoregressive error accumulation
• Outperformed strong RGB-based world-model baselines on RoboTwin in rollout error, semantic similarity, and inference speed
• First author; submitted to ECCV 2026 (under review)

Research Intern, Embodied Multimodal Large Models Research Center

Beijing Academy of Artificial Intelligence (BAAI)

Mentor: Xiaojie Zhang

• Topic: Reinforcement Learning for VLA Models and Flow-Based Robot Policies
• Reproduced representative VLA-RL and robot policy post-training frameworks including RLinf, SimpleVLA-RL, RL4VLA, and DSRL
• Investigated RL objectives for flow-based robot policies, focusing on training stability, inference efficiency, and compatibility with modern VLA backbones
• Contributed to early development of the FORCE project; co-authored an ICML 2026 submission (under review)

Research Intern, Future Chemistry Department

XtalPi

• Topic: Humanoid Robots for Laboratory Automation; VLA Policy Learning and Deployment
• Trained VLA policies for humanoid robot manipulation in laboratory environments, focusing on tool use and structured object interaction
• Deployed policies on real humanoid robot platforms and integrated perception, policy inference, and control for real-world execution
• Developed manipulation capabilities for representative laboratory instruments in automated bio-lab scenarios
• Conducted iterative real-robot evaluation and failure analysis to improve deployment robustness

Research Intern, Embodied Multimodal Large Models Research Center

Beijing Academy of Artificial Intelligence (BAAI)

• Topic: Pretraining Data Pipelines for VLA and Vision-Language Models
• Processed embodied datasets including AgiBot-World and DROID for large-scale VLA pretraining
• Performed action normalization, action tokenization, and dataset reorganization into structured JSON training format
• Contributed to VLM data preprocessing by adding grounding boxes as object-aware visual-language supervision

Raspberry Pi-based Robotic Arm Control

Project Lead

• Developed a Raspberry Pi-based control system for a 3D-printed robotic arm, using GPIO PWM signals to actuate an SG52 servo for gripper control and four-phase eight-step drive signals for 4-phase 5-wire stepper motors at the arm joints
• Implemented two interaction modes: infrared remote control via lirc and camera-based gesture control using MediaPipe for real-time arm manipulation

Desktop Manipulation via 3D Diffusion Policy

Project Lead

• Built and deployed a real-world 3D Diffusion Policy (DP3) pipeline for tabletop manipulation, including multimodal data collection, camera calibration, point-cloud preprocessing, policy training, and online inference
• Validated the system on real tasks including lemon-in-bowl placement, bowl lifting, and drawer pushing using point-cloud observations from a RealSense D435 and an ARX X5 robotic arm

Class Leader, Class 2303

Project Lead

• Led class operations and activities, including event planning, logistics coordination, routine administration, and award application materials
• Class received honors including Excellent Undergraduate Academic Style, Best Sports Team, and Best Popularity Class

Street Dance Club

House Division Lead

• Led the House division by coordinating weekly training with instructors and organizing two campus street dance performances, including rehearsal planning and member coordination