主页

关于我

我是苏黎世大学 Informatics 硕士生，主修人工智能，辅修数据科学。我的研究兴趣集中在 Computer Vision、Robotics 和 Multimodal Models 的交叉方向。

目前，我参与的研究工作包括：

Computer Vision and Geometry Group (ETHZ CVG)：实时 Visual-Inertial-LiDAR Gaussian Splatting SLAM。
Robotics and Perception Group (UZH RPG)：事件相机增强的机器人运动控制策略。

我的经验包括：

机器人与强化学习：机器人手臂和无人机上的 RL/IL 训练与部署。
三维视觉与 SLAM：Gaussian Splatting、Visual Odometry 和 SLAM pipeline。
基础模型：VLA 模型、LLM 和多模态预训练。

此前，我曾在 合合信息科技股份有限公司 担任计算机视觉算法实习生。

欢迎通过 jianwen.cao@uzh.ch 联系我。

论文

[Accepted] Jianwen Cao, et al. Generative Event Pretraining with Foundation Model Alignment. CVPR 2026.
[Accepted, Oral] …, Jianwen Cao, et al. Memory Over Maps: 3D Object Localization Without Reconstruction. IEEE ICRA 2026.
[Published] Jianwen Cao, et al. Fuse Tune: Hierarchical Decoder Towards Efficient Transfer Learning. PRCV 2023.

项目

SO-ARM Reinforcement Learning for Robot Manipulation

05.2026. Built a reinforcement learning pipeline for SO-ARM manipulation, covering simulation setup, policy training, and deployment-oriented evaluation.

Memory Over Maps: 3D Object Localization Without Reconstruction

09.2025 - 03.2026. [Accepted, Oral] IEEE ICRA 2026. Proposed a novel framework for 3D object localization that leverages foundation model memory instead of explicit map reconstruction.

Event Camera-Enhanced Robot Motor Control Policies

02.2025 - 02.2026. Conducted autoregressive pre-training on multimodal data, significantly improving the performance and robustness of downstream vision and control tasks.