Wei Zhan is Co-Director of Berkeley DeepDrive, a leading research center on automotive AI involving many Berkeley faculty and industrial partners. He is also Co-Director of BAIR Center for Humanoid Intelligence, a new center combining leading humanoid robotics research labs towards robotic generalists. He is an Assistant Professional Researcher at UC Berkeley leading a team of Ph.D. students and Postdocs conducting research. He also teaches AI for Autonomy at UC Berkeley.
He is also Chief Scientist of Applied Intuition, a physical AI company delivering autonomy, toolchain and OS to various industrial verticals including passenger/trucking automotive, mining, construction, etc. He leads the research efforts towards next-generation physical AI for various products, and the Research Group of the company.
His research is focused on AI for scalable autonomous systems leveraging robotics, computer vision, machine learning and control techniques to tackle challenges with complex scenes, dynamics and human behavior with applications to autonomous driving and general robotics. He received his Ph.D. degree from UC Berkeley. Four of his publications were awarded in flagship conferences and journals.
Hiring!
Wei Zhan is actively hiring Research Scientists, Research Engineers and Research Interns in Applied Intuition. Apply to the roles in “AI Research” section if you are interested in conducting research on AI for autonomous systems, robotics and simulation, or supporting research on AI infrastructure and software/hardware.
Selected Awards
- Outstanding Paper Award of ICCV 2025 RIWM Workshop
- Best Paper Award in Automation of ICRA 2025
- Best Paper Award of ICRA 2024
- Best Paper Award – Honorable Mention of IEEE Robotics and Automation Letters
- Best Student Paper Award in IV 2018
- 2024 ITS Team Leadership Award from Institute of Transportation Studies at UC Berkeley, “For excellent group leadership and contributions in Berkeley DeepDrive, superb mentoring of PhD research students, and for collaborative excellence”
Selected Research
SPACeR: Self-Play Anchoring with Centralized Reference Models
Wei-Jer Chang, Akshay Rangesh, Kevin Joseph, Matthew Strong, Masayoshi Tomizuka, Yihan Hu, Wei Zhan
ICLR 2026
Paper Website Blogpost
Incorporating behavior imitation into a self-play reinforcement learning framework, enabling human-like, robust and fast reactive behavior generation in large-scale simulation
Learning to Drive is a Free Gift: Large-Scale Label-Free Autonomy Pretraining from Unposed In-The-Wild Videos
Matthew Strong, Wei-Jer Chang, Quentin Herau, Jiezhi Yang, Yihan Hu, Chensheng Peng, Wei Zhan
CVPR 2026
Paper Website
Foundation model pretraining from in-the-wild videos without poses and labels, facilitating planning and various downstream tasks
Selected Projects
Reinforcement Learning, Control, Autonomous Racing
- Residual Q-Learning – policy customization without value: NeurIPS’23, Website, Code
- Residual-MPPI – online policy customization for continuous control: ICLR’25, Website
- Multi-agent RL cost-efficient generalization: RLC’24
- Active exploration for modeling dynamics and racing behavior: IEEE Trans-CST ’24
Generative Model, 3D Reconstruction, Neural Simulation
- X-Drive – Cross-modality Consistent Data Generation with Diffusion: ICLR’25
- DeSiRe-GS – 4D Gaussians for Decomposition and Mesh: CVPR’25
- CompGS – Compositional Text-to-3D Gaussians: CVPR’25
- Q-SLAM – quadric representations for monocular SLAM: CoRL’24
- S3 Gaussian – Self-Supervised Street Gaussian: arxiv, Code
Manipulation, Diffusion Policy, Robot Learning from Human
- Generalizable representation learning human demonstrations: RSS’24, Website
- Open X-Embodiment – Robotic Learning Datasets and RT-X Models: ICRA’24 (Best Paper Award), Blog, Dataset, Website, Code
- Sparse Diffusion Policy – Flexible Policy with Mixture of Experts (MoE): CoRL’24
- DexHandDiff – Interaction-aware Diffusion for Adaptive Manipulation: CVPR’25
- PhyGrasp – grasping with physics-informed large models: IROS’25, Website
3D Perception, Fusion, Data Engine
- SOLOFusion – temporal multi-view 3D detection: ICLR’23 (notable top 5% oral presentation), Code
- SparseFusion – fusing multi-modal sparse representations: ICCV’23, Code
- Free Data selection with general-purpose models: NeurIPS’23
- Cross-modality semi-supervised learning for 3D detection: ECCV’22, Code
Behavior Generation, Language Reasoning, Diagnosis
- WOMD-Reasoning – language Dataset for interaction reasoning: ICML’25, Website
- LANGTRAJ: language-conditioned generation model and dataset: ICCV’25
- Efficient Diffusion Models for Prediction and Controllable Generation: ECCV’24
- Code diagnosis and repair of motion planners by LLM: RA-Letters’24
- Guided diffusion for traffic simulation with controllable criticality: ECCV’24
Prediction, INTERACTION Dataset and Benchmark
- INTERACTION dataset with critical scenes and densely interactive behavior: Website, arxiv, INTERPRET challenge
- Scenario-transferable with semantic intention representation: IEEE Tran-ITS ’22, Video summary, IV’18 (Best Student Paper Award)
- Multi-agent prediction combining egocentric and allocentric views: CoRL’21
- Social posterior collapse in variational autoencoder: NeurIPS’21
Planning, Behavior Design, Inverse Reinforcement Learning
- Socially compatible planner: RA-Letters ’20 (Best Paper Award – Honorable Mention)
- Constrained Iterative LQR/LQG: IEEE Trans-IV ’19
- Imitation learning trained and combined with model predictive control: DSCC’18
- Probabilistic prediction with hierarchical inverse reinforcement learning: ITSC’18