CV
Education, work experience, technical skills, and academic service.
01
General Information
| Full Name | Peizheng Li |
| peizheng.li@yahoo.com | |
| Homepage | edwardleelpz.github.io |
| Languages | Chinese (native), English (professional), German (professional) |
02
Education
-
2023 - Present Ph.D. in Computer Science
University of Tübingen, Germany - Advisors: Prof. Andreas Geiger and Prof. Andreas Zell
- Research: multimodal learning, VLM / VLA, world models, embodied AI, robotics, and autonomous driving
-
2020 - 2022 M.Sc. in Electromobility
University of Stuttgart, Germany - Grade: 1.6 (German scale, 1.0 best; approx. 3.7 / 4.0 US equivalent)
- Thesis: End-to-End Agent Perception and Occupancy-Based Motion Prediction in Bird's-Eye View
-
2014 - 2019 B.Eng. in Vehicle Engineering
Tongji University, Shanghai, China - Grade: 4.5 / 5.0
- Thesis: Vehicle Detection and Tracking Based on Sensor Fusion
03
Work Experience
-
2023 - Present Industrial Ph.D. Researcher
Mercedes-Benz Group AG, Stuttgart, Germany Scene Understanding Group, R&D - Research on foundation models for autonomous systems, spanning VLM / VLA for autonomous driving, open-world 3D occupancy prediction, spatial and world-model-oriented reasoning, diffusion-based motion planning, self-supervised scene flow, and robotics / HRI collaborations.
- First-author publications: SpaceDrive (CVPR 2026), AGO (ICCV 2025), PowerBEV (IJCAI 2023).
- Built large-scale data-engine and evaluation pipelines for data acquisition, cleaning, temporal consistency, scene / instance retrieval, and automated dense 3D pseudo-label generation.
- Scaled training and evaluation across local GPU clusters and cloud using PyTorch, CUDA, DDP, Docker, Kubernetes, Flyte, Azure, and GCP.
- Mentored interns and master thesis students; collaborated across research and engineering teams on open-source releases and reproducible evaluation.
-
2022 Master Student
Mercedes-Benz Group AG, Stuttgart, Germany Scene Understanding Group, R&D - Built a camera-based end-to-end BEV perception and future prediction model with Lift-Splat-Shoot-style view transformation, temporal prediction, and multi-task decoding; achieved 39.3% dynamic IoU on nuScenes.
- Designed a multi-stage temporal GCN for the Waymo Occupancy and Flow Prediction Challenge; achieved 74.2% Flow-Grounded Occupancy AUC and ranked 4th on the leaderboard.
-
2021 Research Intern
Mercedes-Benz Group AG, Stuttgart, Germany Scene Understanding Group, R&D - Studied contextual bias in 2D object detection and domain adaptation; disruptive context reduced mAP by 5.37%.
- Developed context-separation components and baseline pipelines to support internal adaptation research.
04
Technical Skills
| Programming | |||||||
| Python | |||||||
| C++ | |||||||
| CUDA | |||||||
| C# | |||||||
| MATLAB | |||||||
| Bash | |||||||
| ML / Systems | |||||||||||
| PyTorch | |||||||||||
| TensorFlow | |||||||||||
| OpenMMLab | |||||||||||
| Distributed Training | |||||||||||
| Docker | |||||||||||
| Kubernetes | |||||||||||
| Flyte | |||||||||||
| W&B | |||||||||||
| Azure | |||||||||||
| GCP | |||||||||||
| Research Areas | ||||||||||||||
| Foundation Models & VLM / VLA for Driving | ||||||||||||||
| World Models | ||||||||||||||
| Multimodal Learning | ||||||||||||||
| 3D Vision | ||||||||||||||
| Open-World Perception | ||||||||||||||
| BEV / Occupancy | ||||||||||||||
| Scene Flow | ||||||||||||||
| Auto-Labeling & Data Engines | ||||||||||||||
| Spatial Intelligence | ||||||||||||||
| Embodied AI | ||||||||||||||
| Multi-View Geometry | ||||||||||||||
| SLAM | ||||||||||||||
| HRI | ||||||||||||||
05
Service
- Organizer: 4th DriveX Workshop, in conjunction with CVPR 2026
- Organizer: 6th DriveX Workshop, in conjunction with ECCV 2026
- Reviewer: CVPR 2025/2026, ICCV 2025, ECCV 2026, ICRA 2026, AAAI 2026/2027, IROS 2025/2026, IEEE T-ITS 2025
CV