Autonomous Driving
End-to-end trajectory planning, navigation-conditioned behavior, and real-time embedded deployment.
Research Scientist - San Jose, CA
Zekai Yin
Research Scientist working on autonomous driving, multimodal planning, visual odometry, and embodied AI.
I build learning systems for trajectory planning and perception: Flow Matching and diffusion-style planners, multimodal prediction, visual odometry, robotics, and photorealistic 3D simulation.
About
Planning and perception for embodied autonomy.
I am a Research Scientist, Deep Learning Algorithm at Black Sesame Technologies, where I research and engineer end-to-end multimodal trajectory planning models for autonomous driving.
Previously, I worked with the H2X Lab at Boston University on ZeroVO, multimodal driving decision models, 3D reconstruction-based simulation, and guide-dog navigation datasets. Before BU, I worked with PKU-Agibot Lab on camera-to-robot pose estimation and robot manipulation. I hold an M.S. in Artificial Intelligence from Boston University and a B.S. in Data Science and Big Data Technology from Peking University.
Generative Planning
Flow Matching, diffusion models, trajectory tokenization, anchor scoring, and multimodal futures.
Visual Odometry
Calibration-light geometry, VLM features, cross-attention, and synthetic driving data.
Robotics and 3D Simulation
Pose estimation, ROS systems, NeRF, Gaussian Splatting, VR studies, and embodied AI datasets.
Publications
Accepted papers.
* indicates equal contribution.
CVPR 2023
CVPR 2025
CoRL 2025
Selected projects
Research systems across planning, perception, and simulation.
Hover to play
CVPR 2025
ZeroVO (Visual Odometry)
Visual odometry with minimal assumptions, fusing VLM features and geometric information through multi-head cross-attention.
Experience
Applied research from lab prototypes to deployment.
Aug 2025 - Present
Research Scientist, Deep Learning Algorithm
Black Sesame Technologies - San Jose, CA
- Research-engineered an end-to-end multimodal Flow-Matching trajectory planning model with VQ-VAE and RFSQ trajectory tokens.
- Built diffusion-likelihood anchor scoring, goal-based heatmap-offset agent prediction, and embedded ONNX deployment paths.
- Developed navigation-conditioned prediction and planning components for challenging lane-change, cut-in, crossing, and intrusion scenarios.
Jan 2024 - Aug 2025
Machine Learning Engineer / Research Assistant
H2X Lab, Boston University
- Proposed ZeroVO as co-first author, fusing VLM features and geometry with multi-head cross-attention for CVPR 2025.
- Created a GTA synthetic driving dataset with 300,000 images across 1,200 videos.
- Proposed a GMM-based diffusion model for multimodal human-like driving behavior, accepted by CoRL 2025.
- Developed VR-integrated simulation with 3D Gaussian Splatting and NeRF, collecting 20,000+ trajectories.
Jul 2022 - May 2023
Machine Learning Engineer / Research Assistant
PKU-Agibot Lab, Peking University
- Proposed Structure-Guided Temporal Attention Pose as co-first author for CVPR 2023.
- Built a Blender synthetic dataset with 180,000 images for camera-to-robot pose estimation.
- Worked with ROS, PyBullet, Franka Panda, SAM, 6D pose estimation, and XARM6 robotic manipulation pipelines.
Education
Boston University
M.S. in Artificial Intelligence, Sep 2023 - Jan 2025
Peking University
B.S. in Data Science and Big Data Technology, Sep 2019 - Jul 2023
Leadership
Hands-on teaching, lab building, and maker education.
Mar 2020 - Jul 2023
Tutor and Course Organizer
Yuanpei College Woodwork Class, Peking University
- Founded a carpentry course for engineering education, developing hands-on curriculum and safety protocols for students.
- Expanded the program into a co-cultivation initiative between Yuanpei College and Beijing 101 Middle School.
- Instructed 150+ students over three years and received the 2022 YuanPei Special Contribution Award scholarship.
Feb 2023 - Jul 2023
Founder and Leader
Yuanpei College 3D Printing and Designing Lab
- Established the college's first 3D printing lab, implementing CAD-to-fabrication workflows with material printing capabilities.
- Designed graduation gifts using parametric modeling techniques and additive manufacturing production methods.
Skills
Tools and systems I use.
Programming
Python, C++, C, MATLAB
ML Frameworks
PyTorch, TensorFlow, Scikit-learn, JAX, Keras, HuggingFace
ML Engineering
MLflow, Weights & Biases, Docker, Kubernetes, ONNX, TensorRT, Ray
Computer Vision
OpenCV, Detectron2, YOLO, SAM, NeRF, 3D Reconstruction, Gaussian Splatting
Robotics and Simulation
ROS, ROS2, Blender, PyBullet, Libfranka, Franka-Control, CAD, Fusion 360
Data Processing
NumPy, Pandas, Matplotlib, SciPy, Dask, Spark, Luigi
Contact
Research collaborations, planning systems, and autonomy roles.
I am based in San Jose, CA. The fastest way to reach me is email.