Wenzhao Zheng

I am currently a postdoctoral fellow in the Department of EECS at University of California, Berkeley, affiliated with Berkeley Artificial Intelligence Research Lab (BAIR) and Berkeley Deep Drive (BDD) , supervised by Prof. Kurt Keutzer . Prior to that, I received my Ph.D degree from the Department of Automation at Tsinghua University, advised by Prof. Jie Zhou and Prof. Jiwen Lu . In 2018, I received my BS degree from the Department of Physics, Tsinghua University.

I am generally interested in artificial intelligence and deep learning. My current research focuses on:

🦙 Large Models + 🚙 Embodied Agents -> 🤖 AGI

Email  /  CV  /  Google Scholar  /  GitHub

*Equal contribution    ^†Project leader/Corresponding author.

Newest Papers

GaussianToken represents each image by a set of 2D Gaussian tokens in a feed-forward manner.

UniDrive presents the first comprehensive framework designed to generalize vision-centric 3D perception models across diverse camera configurations.

GaussianFormer-2 interprets each Gaussian as a probability distribution of its neighborhood being occupied and conforms to probabilistic multiplication to derive the overall geometry.

GlobalMamba constructs a causal token sequence by frequency, while ensuring that tokens acquire global feature information.

Visual 2-Dimensional Mamba (V2M) generalize SSM to the 2-dimensional space and generates the next state considering two adjacent states on both dimensions (e.g., columns and rows) which directly processes image tokens in the 2D space.

Selected Papers [Full List]

Owl-1 approaches consistent long video generation with an omni world model, which models the evolution of the underlying world with latent state, explicit observation and world dynamics variables.

OccWorld models the joint evolutions of 3D scenes and ego movements and paves the way for interpretable end-to-end large driving models.

SparseVLM sparsifies visual tokens adaptively based on the question prompt.

Doe-1 is the first closed-loop autonomous driving model for unified perception, prediction, and planning.

GPD-1 proposes a unified approach that seamlessly accomplishes multiple aspects of scene evolution, including scene simulation, traffic simulation, closed-loop simulation, map prediction, and motion planning, all without additional fine-tuning.

Spatial-Temporal simulAtion for drivinG (Stag-1) enables controllable 4D autonomous driving simulation with spatial-temporal decoupling.

Driv3R predicts per-frame pointmaps in the global consistent coordinate system in an optimization-free manner.

GaussianAD is a Gaussian-centric end-to-end framework which employs sparse yet comprehensive 3D Gaussians to pass information through the pipeline to efficiently preserve more details.

GenAD casts end-to-end autonomous driving as a generative modeling problem.

GaussianWorld reformulates 3D occupancy prediction as a 4D occupancy forecasting problem conditioned on the current sensor input and propose a Gaussian World Model to exploit the scene evolution for perception.

GaussianFormer proposes the 3D semantic Gaussians as a more efficient object-centric representation for driving scenes compared with 3D occupancy.

SelfOcc is the first self-supervised work that produces reasonable 3D occupancy for surround cameras.

Given only surround-camera motorcycle RGB images barrier as inputs, our model (trained using trailer only sparse traffic cone LiDAR point supervision) can predict the semantic occupancy for all volumes in the 3D space.

EmbodiedOcc formulates an embodied 3D occupancy prediction task and employs a Gaussian-based framework to accomplish it.

PixelGaussian dynamically adjusts the Gaussian distributions based on geometric complexity in a feed-forward framework.

S³Gaussian employs 3D Gaussians to model dynamic scenes for autonomous driving without other supervisions (e.g., 3D bounding boxes).

Website Template