PhD Researcher | Spatial Intelligence · Embodied AI · Foundation Models for Physical World
University of Tübingen ·
Mercedes-Benz AG R&D
My research focuses on building autonomous systems that perceive, model, and reliably act in the 3D physical world, spanning 3D/4D dynamics scene understanding, open-world modeling, spatial intelligence, and embodied systems.
I am a PhD researcher at the University of Tübingen and Mercedes-Benz AG R&D, advised by Prof. Andreas Geiger and Prof. Andreas Zell. My work bridges academic research and industrial-scale engineering, spanning large-scale data pipelines and real-world vehicle deployments, with publications at CVPR, ICCV, ECCV, ICML, IJCAI, and IROS as first or core author.
Getting AI into the physical world is pretty cool I guess.
- 3D World Modeling — 3D/4D physical scene modeling that encode geometry, semantics, and dynamics for open world
- Spatial Intelligence for VLM/VLA — spatial awareness into VLM/VLA and physically-consistent 3D reasoning
- Self-Supervised 3D Dynamics — scene flow estimation, motion understanding, and scalable learning from real-world data
- Embodied Robotics — foundation-model-driven multimodal interaction for robot-human-collaboration in physical environments
SpaceDrive: Infusing Spatial Awareness into VLM-based Autonomous Driving
Infusing explicit spatial representations into vision-language models for robust autonomous driving with 3D spatial reasoning.
Peizheng Li, Zhenghao Zhang, David Holtz, Hang Yu, Yutong Yang, Yuzhi Lai, Rui Song, Andreas Geiger, Andreas Zell
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026) · 🏆 #1 on nuScenes · #2 on Bench2Drive
AGO: Adaptive Grounding for Open World 3D Occupancy Prediction
Adaptive grounding framework that bridges 2D vision-language features to open-world 3D occupancy prediction without manual vocabulary.
Peizheng Li, Shuxiao Ding, You Zhou, Qingwen Zhang, Onat Inak, Larissa Triess, Niklas Hanselmann, Marius Cordts, Andreas Zell
International Conference on Computer Vision (ICCV 2025)
SeFlow: A Self-Supervised Scene Flow Method in Autonomous Driving
Self-supervised scene flow estimation from point clouds, eliminating the need for expensive human annotations.
Qingwen Zhang, Yi Yang, Peizheng Li, Olov Andersson, Patric Jensfelt
European Conference on Computer Vision (ECCV 2024) · 🏆 #1 on Argoverse 2 Self-Supervised Scene Flow leaderboard
PowerBEV: A Powerful Yet Lightweight Framework for Instance Prediction in Bird's-Eye View
Lightweight yet powerful BEV framework for joint instance segmentation and future motion prediction.
Peizheng Li, Shuxiao Ding, Xieyuanli Chen, Niklas Hanselmann, Marius Cordts, Juergen Gall
International Joint Conference on Artificial Intelligence (IJCAI 2023)
G2DP: Diffusion Planning with Spatio-Temporal Grid Guidance
Grid-guided diffusion planner injecting dense spatio-temporal cost gradients into denoising for safe, route-adherent closed-loop driving.
Hang Yu, Ye Jin, Alessandro Canevaro, Julian Schmidt, Julian Jordan, Peizheng Li, Marc Kaufeld, Silvan Lindner, Johannes Betz, Wilhelm Stork
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2026) · 🏆 SOTA on nuPlan (closed-loop)
Shift & Drift: A Zero-Shot Benchmark for Generalizable and Robust Autonomous Driving Motion Planning
Zero-shot dual-track benchmark stress-testing motion planners under semantic shift and state-distribution drift in closed-loop driving.
Alessandro Canevaro, Hang Yu, Julian Schmidt, Peizheng Li, Silvan Lindner, Wilhelm Stork, Georg Martius, Julian Jordan
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2026)
FAM-HRI: Foundation-Model Assisted Multimodal Human-Robot Interaction Combining Gaze and Speech
Foundation-model assisted multimodal HRI that fuses gaze and speech via LLMs for intuitive robot manipulation.
Yuzhi Lai, Shenghai Yuan, Peizheng Li, Boya Zhang, Benjamin Kiefer, Tianchen Deng, Andreas Zell
IEEE Transactions on Automation Science and Engineering (T-ASE 2026)
Seizure-Semiology-Suite (S3): A Clinically Multimodal Dataset, Benchmark, and Models for Seizure Semiology Understanding
Clinically grounded multimodal benchmark and models for fine-grained seizure semiology understanding.
Lina Zhang, Tonmoy Monsoor, Peizheng Li, et al.
International Conference on Machine Learning (ICML 2026, Highlight)
Can Multimodal Large Language Models Understand Pathologic Movements? A Pilot Study on Seizure Semiology
Pilot study evaluating multimodal large language models for interpretable pathological movement recognition in seizure videos.
Lina Zhang, Tonmoy Monsoor, Mehmet Efe Lorasdagi, Prateik Sinha, Chong Han, Peizheng Li, et al.
IEEE Engineering in Medicine and Biology Society (EMBC 2026)
Glance-Say: Multimodal Human-Robot Collaboration and Intent Recognition via Sticky Glance
Gaze-speech multimodal interaction fusing sticky-glance intent stabilization and shared control for assistive robotic manipulation.
Yuzhi Lai, Shenghai Yuan, Peizheng Li, Benjamin Kiefer, Andreas Zell
arXiv:2603.06121, 2026
TQD-Track: Temporal Query Denoising for 3D Multi-Object Tracking
Temporal query denoising approach for robust 3D multi-object tracking in autonomous driving scenarios.
Shuxiao Ding, Yutong Yang, Julian Wiederer, Markus Braun, Peizheng Li, Juergen Gall, Bin Yang
arXiv:2504.03258, 2025
SEER-VAR: Semantic Egocentric Environment Reasoner for Vehicle Augmented Reality
Semantic egocentric environment reasoning for vehicle augmented reality with spatial understanding.
Yuzhi Lai, Shenghai Yuan, Peizheng Li, Jun Lou, Andreas Zell
arXiv:2508.17255, 2025
- Jul 2026 — G2DP and Shift & Drift accepted at IROS 2026 🎉
- Feb 2026 — SpaceDrive accepted at CVPR 2026 🎉 · #1 nuScenes open-loop · #2 Bench2Drive closed-loop
- Jun 2025 — AGO accepted at ICCV 2025 🎉
- Jul 2024 — SeFlow accepted at ECCV 2024 🎉 · #1 Argoverse 2 self-supervised scene flow
- Apr 2023 — PowerBEV accepted at IJCAI 2023 🎉
I am open to research scientist / research engineer positions in spatial AI, robotics, autonomous driving, and embodied intelligence.
