Research Fellow · Nanyang Technological University
Lecturer · Yanshan University
I am a Research Fellow at Nanyang Technological University, working with Prof. Chen Lv, and a Lecturer and Master's Supervisor at the School of Artificial Intelligence, Yanshan University.
I received my Ph.D. in Computer Science and Technology from Beijing Jiaotong University in 2026, advised by Prof. Caiyan Jia. During my doctoral studies, I also conducted joint research at Tsinghua University.
My research focuses on world models, embodied intelligence, and autonomous driving. I am interested in how predictive models and generative planning can turn multimodal observations into reliable actions in the physical world.
Predictive representations, physical dynamics, and spatial intelligence for embodied agents.
End-to-end driving, flow matching, diffusion models, and reinforcement learning for planning.
LiDAR–camera fusion, 3D object detection, BEV feature alignment, and robustness.
Recent manuscripts and surveys.
Preprint, 2026
A benchmark and evaluation framework for failure recovery in robotic manipulation after failed grasps, collisions, and unintended object movements.
Preprint, 2026
Conditional flow matching predicts sequences of future latent states for robust, reconstruction-free JEPA world models.
Preprint, 2026
A systematic survey of diffusion-based world models, covering methods, applications, datasets, characteristics, and open challenges.
* Corresponding author.
NeurIPS 2026 · Accepted
Future-aware latent representations condition diffusion-based trajectory planning for autonomous driving.
CVPR 2026Highlight
Constraint-guided flow matching for diverse, physically feasible driving trajectories.
IEEE TPAMI, 2026 · Early Access
Reinforcement learning guides diffusion toward diverse and safe end-to-end driving trajectories.
International Journal of Computer Vision (IJCV), 2026 · Accepted
A unified alignment framework for robust multimodal perception and end-to-end driving.
SCIENTIA SINICA Informationis (中国科学：信息科学), 2026 · Accepted
Visual foundation models for robust multimodal 3D detection. 基于视觉大模型的鲁棒多模态3D检测方法。
IEEE TCSVT, 2026
Dual-guided instance fusion improves robust multimodal 3D detection of distant, small and occluded objects.
Automotive Innovation, 2026
Local agent interactions and motion-aware training improve end-to-end autonomous driving planning.
CVPR 2025
Trajectory and perception momentum improve the temporal consistency of driving plans.
Scientific Data, 2025
A real-world sensor corruption benchmark evaluates the robustness of autonomous driving perception.
ECCV 2024
Local and global alignment strengthen LiDAR–camera fusion under calibration errors.
IJCAI 2024
Visual foundation model priors improve robust multimodal 3D perception.
IEEE TITS, 2024
A systematic view of accuracy, latency and robustness in autonomous driving perception.
IEEE TCSVT, 2024
Graph matching improves feature alignment between heterogeneous sensor modalities.
ICCV 2023
Graph-based correspondence addresses LiDAR–camera feature misalignment.
IEEE TGRS, 2023
Voxel fusion connects sparse point clouds and dense image features for 3D detection.
IEEE TGRS, 2023
Treating voxels as points for LiDAR-based 3D object detection.
Research Fellow · Advisor: Prof. Chen Lv
World models, embodied intelligence, and autonomous driving.
Lecturer & Master's Supervisor · School of Artificial Intelligence
Computer Science and Technology · Advisor: Prof. Caiyan Jia
Dissertation: Robustness optimization methods for autonomous driving systems.
Tsinghua–DiDi Joint Research Center for Future Mobility & Lion Autonomous Driving Laboratory.
Computer Technology · Advisor: Prof. Kuihe Yang
Network Engineering
Technological Invention Award · Third-ranked contributor
CICC 科学技术奖 · 技术发明奖二等奖（排名第三）
For research discussions and collaboration, please reach me at 22110110@bjtu.edu.cn.