Zhao Yang (杨昭)

I am pursuing my PhD in the School of Artificial Intelligence at Xi’an Jiaotong University (XJTU),
with research affiliations at the State Key Laboratory of Human-Machine Hybrid Augmented Intelligence,
the National Engineering Research Center for Visual Information and Applications,
and the Institute of Artificial Intelligence and Robotics.

My research lies at the intersection of world-model-centric embodied AI, autonomous driving, and vision-language-action (VLA) systems.
I’m particularly interested in how latent world models, long-horizon control, and diffusion-based planning can be combined to make robot policies reliable, temporally consistent, and deployable on real platforms.

Previously, I worked on world-model and perception systems at Baidu Apollo (ADFM), Alibaba DAMO Academy, and Huawei Noah’s Ark Lab, and I am now founding an embodied-intelligence startup, Cytoderm Intelligent Technology.

Research Overview

My long-term goal is to build world-model-first autonomous robotics capable of robust long-horizon reasoning, dexterous manipulation, and real-world deployment.

Methodologically, my work spans:

• World Models & Generative Planning
  Latent 3D / video world models, diffusion-based planners, and chunked action policies for long-horizon control in manipulation and driving.

• Vision-Language-Action (VLA) for Robotics
  World-model-centric VLA architectures for generalizable robotic behavior, including dexterous grasping and manipulation tasks with action chunking, seam-aware execution, and RL fine-tuning over large-scale, unstructured robotic data.

• Autonomous Driving & 3D Perception
  Semi-supervised BEV 3D detection, panoramic surround-view sensing, uncertainty-aware localization, and HD map construction for embodied navigation.

On the application side, my work has been validated on large-scale robotic manipulation benchmarks such as CALVIN, LIBERO, and real-world robotic platforms, as well as autonomous-driving datasets including nuScenes, KITTI-360, and Waymo.

Career & Education History

Expertise

• Embodied AI & Robot Learning (2025 – Present)
  World models, VLA policies, diffusion-based planners, chunked and hierarchical control.

• Autonomous Driving & 3D Perception (2020 – 2025)
  BEV 3D detection, surround-view fisheye perception, HD map construction, planning under uncertainty.

• Computer Vision & Machine Learning (2013 – 2019)
  Semi-supervised learning, tensor methods, model compression and distillation.

News

• 2025.12 – Released Cytoderm·Robotic Arm, a lightweight robotic manipulation platform designed for embodied AI research and industrial prototyping. More details: https://www.cytoderm.ai/
• 2025.12 – Launched Cybopal, a consumer-level embodied-intelligence robot product for everyday interactive and assistive tasks. See product page: https://cybopal.com/
• 2025.11 – ChunkFlow: Towards Continuity-Consistent Chunked Policy Learning completed and under review at a top venue in robot learning / AI.
• 2025.10 – Object-Guided Semi-Supervised BEV 3D Object Detection with 3D Box Refinement accepted by IEEE T-ITS.
• 2025.09 – Finished large-scale experiments on LIBERO long-horizon manipulation benchmark with world-model-centric VLA agents.
• 2024.06 – 1st-place solution for CVPR 2024 Autonomous Driving Grand Challenge – Predictive World Model Track.
• 2024.03 – Multiple works on BEV detection, spherical fusion, and map construction accepted to CVPR / ACM MM.

Publications

A more complete and up-to-date list is available on Google Scholar.

Recent works (2025–2024)

Under review

ChunkFlow: Towards Continuity-Consistent Chunked Policy Learning

Zhao Yang, Yinan Shi, et al.

Under review, 2025.

VLA policy with chunked actions, overlap blending, and continuity-constrained RL to suppress boundary jitter in long-horizon control.

[PDF] [Code] [Project]

ACM MM 2025

DriVerse: Navigation World Model for Driving Simulation via Multimodal Trajectory Prompting and Motion Alignment

Xiaofan Li, Chenming Wu, Zhao Yang*, et al.

ACM Multimedia (ACM MM), 2025.

A navigation-centric world model that conditions on multimodal trajectory prompts and enforces motion-aligned latent dynamics for driving simulation.

[PDF] [Code]

ICCV 2025

U-ViLAR: Uncertainty-Aware Visual Localization for Autonomous Driving via Differentiable Association and Registration

Xiaofan Li, Zhihao Xu, Chenming Wu, Zhao Yang, et al.

IEEE/CVF International Conference on Computer Vision (ICCV), 2025.

Differentiable association and registration with explicit uncertainty modeling for robust large-scale localization in autonomous driving.

[PDF]

IROS 2025

Causal-Planner: Causal Interaction Disentangling with Episodic Memory Gating for Autonomous Planning

Yibo Yuan, Jianwu Fang, Yang Zhou, Zhao Yang, et al.

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025.

Causal interaction modeling and memory-gated policy for interpretable planning in dynamic driving environments.

[PDF] [Code]

ICRA 2025

Towards Accurate Semi-Supervised BEV 3D Object Detection with Depth-Aware Refinement and Denoising-Aided Alignment

Zhao Yang, Yinan Shi, et al.

IEEE International Conference on Robotics and Automation (ICRA), 2025.

Semi-supervised BEV detector with depth-aware refinement, denoising-aided alignment, and robust pseudo-labeling on nuScenes.

[PDF] [Code]

Under Review

DualDiff+: Dual-branch Diffusion for High-fidelity Video Generation with Reward Guidance

Zhao Yang, Zezhong Qian, et al.

Under review, 2025.

Dual-branch diffusion with semantic and motion reward alignment for high-resolution, temporally coherent video synthesis.

[PDF] [Code] [Project]

ICRA 2025

DualDiff: Dual-Branch Diffusion Model for Autonomous Driving with Semantic Fusion

Zhao Yang, Haoteng Li, et al.

IEEE International Conference on Robotics and Automation (ICRA), 2025.

Dual-branch diffusion model that fuses semantic BEV priors with image features for controllable driving video and occupancy generation.

[PDF] [Code] [Project]

T-ITS 2025

Object-Guided Semi-Supervised Bird’s-Eye View 3D Object Detection with 3D Box Refinement

Zhao Yang, Yinan Shi, et al.

IEEE Transactions on Intelligent Transportation Systems (T-ITS), 2025.

Semi-supervised BEV framework with object-guided consistency and box refinement, improving label efficiency on large-scale driving datasets.

[PDF] [Code]

Info Fusion 2025

EquivFisheye: A Spherical Fusion Framework for Panoramic 3D Perception with Surround-View Fisheye Cameras

Zhao Yang, Xinglin Pu, et al.

Information Fusion, 2025.

Spherical feature fusion and equivariant pooling for efficient panoramic 3D perception from multi-view fisheye cameras.

[PDF] [Code]

CVPR 2024

Cadkp: Category-aware Knowledge Distillation and Pruning Framework for Lightweight 3D Object Detection

Haonan Zhang, Longjun Liu, Yuqi Huang, Zhao Yang, et al.

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024.

Category-aware distillation and structured pruning to obtain compact yet accurate 3D detectors for autonomous driving.

[PDF] [Code]

ACM MM 2024

RAG-Guided Large Language Models for Visual Spatial Description with Adaptive Hallucination Correction

Jun Yu, Yunxiang Zhang, Zerui Zhang, Zhao Yang et al.

ACM Multimedia (ACM MM), 2024.

Retrieval-augmented LLMs for spatial description, with adaptive hallucination correction in vision-language reasoning.

[PDF]

ACM MM 2024

Temporal-Informative Adapters in VideoMAE V2 and Multi-Scale Feature Fusion for Micro-Expression Spotting-then-Recognize

Jun Yu,Guopeng Zhao, Yaohui Zhang, Peng He, Zerui Zhang, Zhao Yang, et al.

ACM Multimedia (ACM MM), 2024.

Temporal adapters and multi-scale fusion for fine-grained micro-expression analysis.

[PDF]

ACM MM 2024

IC-Mapper: Instance-Centric Spatio-Temporal Modeling for Online Vectorized Map Construction

Jiangtong Zhu*, Zhao Yang*, et al.

ACM Multimedia (ACM MM), 2024.

Instance-centric spatio-temporal modeling for online HD map vectorization from autonomous driving logs.

[PDF] [Code]

CVPRW 2024 · 1st place

The 1st-Place Solution for CVPR 2024 Autonomous Driving Grand Challenge Track on Predictive World Model

Z. Yang et al.

CVPR 2024 Workshop on Autonomous Driving, 2024.

Large-scale world-model solution for multi-step predictive planning in autonomous driving, achieving 1st place in the CVPR 2024 challenge.

[Project]

Earlier works

Journal 2022

Flexible Android Malware Detection Model Based on Generative Adversarial Networks with Code Tensor

Zhao Yang, Fengyang Deng, et al.

IEEE Transactions on Cyber-Enabled Distributed Computing and Systems, 2022.

GAN-based code-tensor modeling for robust Android malware detection.

[Paper]

T-CSS 2020

Secure Tensor Decomposition for Heterogeneous Multimedia Data in Cloud Computing

Zhao Yang, Cai Fu, et al.

IEEE Transactions on Computational Social Systems (T-CSS), 2020.

Privacy-preserving tensor decomposition framework for heterogeneous multimedia data in cloud environments.

[Paper]

CVPRW 2020 · 1st place

The 1st-Place Solution for WebVision CVPR 2020 Virtual Challenge

Zhao Yang et al.

CVPR WebVision Challenge Workshop, 2020.

Large-scale web-vision model with robust training under noisy labels, achieving 1st place in the WebVision 2020 challenge.

[Project]

Service & Misc.

• Reviewer for major conferences and journals in vision and robotics (CVPR, ICCV, ECCV, ACM MM, ICRA, IROS, T-ITS, etc.).
• Extensive experience with large-scale GPU clusters (A800 / A100), distributed training, and real autonomous driving platforms.

If you are interested in collaborations on world models, VLA agents, or diffusion-based planning, feel free to contact me by email.