Wenyuan Xie 谢文远
I am a third-year master's student at Shanghai Jiao Tong University's Paris Elite Institute of Engineering, preparing applications for 2027 Fall. My research studies vision-language-action models, embodied navigation, robot manipulation, and post-training methods that make intelligent agents more reliable.
I completed both my bachelor's and master's studies at the Paris Elite Institute of Engineering, and I enjoy fitness and thoughtful conversations about robots and learning.
Publications & projects
My research connects structured geometric representations with learning-based agents, so they can adapt from experience and execute long-horizon tasks robustly.
MVP-Nav: Multi-layer Value Map Planner Navigator
A multi-layer value-map planner that combines VGGT, Grounded-SAM and vision-language models for robust 3D navigation.
Navi-Agent: Unlocalized Monocular Navigation Agent
A coordinate-free visual anchor graph for place recognition, progress verification, and recovery in continuous environments.
RelAfford6D: Relational 6D Affordance Graphs
Language-grounded relational affordances become kinematic constraints and closed-loop SE(3) trajectories for articulated-object manipulation.
Dejavu: Towards Experience Feedback Learning
An experience feedback network augments a frozen VLA policy with retrieved trajectories, enabling post-deployment learning without rewriting the base model.
Recovering Hidden Reward in Diffusion-Based Policies
Learning reward signals hidden inside diffusion-policy behavior to improve reliable action generation.
TRUST: Uncertainty-Aligned Tool-Calling Decisions
Reinforcement learning with uncertainty-aware rewards for more reliable decisions in multi-turn agent tool use.
Education
2024.09 - present - Master's student (third year), Paris Elite Institute of Engineering, Shanghai Jiao Tong University - preparing 2027 Fall applications
2020.09 - 2024.06 - Bachelor's student, Paris Elite Institute of Engineering, Shanghai Jiao Tong University
Before 2020 - Hangzhou Xuejun High School
Internships
2026.03 - 2026.07 - Research intern, Agibot Robotics - reinforcement learning post-training for long-horizon phone packaging and whole-body control for wheeled-legged robots
2024.06 - 2024.09 - Research intern, Alibaba Cloud - visual navigation fine-tuning with ViNT and NoMaD; addressed straight-motion bias with turning-data augmentation and a turn-misclassification penalty
2022.06 - 2022.08 - Product intern, Alibaba Cloud - product carbon-footprint certification for the Hangzhou Asian Games mascot plush toy, the Games’ first zero-carbon licensed product; supported the China Academy of Art low-carbon platform
Skills
Tools: Python, PyTorch, JAX, C++, Java, MATLAB · Topics: VLA, VLN, agents, world models, reinforcement learning
谢文远 Wenyuan Xie
我目前是上海交通大学巴黎卓越工程师学院电子信息硕士三年级学生，准备申请 2027 Fall。研究方向包括视觉语言动作模型、具身导航、机器人操作、世界模型和智能体后训练。
论文与项目
研究聚焦将视觉表征、物理几何和学习型智能体结合，使系统能够从经验中适应并完成长程任务。
MVP-Nav: Multi-layer Value Map Planner Navigator
结合 VGGT、Grounded-SAM 与 Multi-layer Value Map，解决纯视觉导航中语义目标与物理约束的结合问题。
Navi-Agent: Unlocalized Monocular Navigation Agent
通过 Visual Anchor Graph 解决无定位条件下导航系统的移动与位置认知问题。
RelAfford6D: Relational 6D Affordance Graphs
利用关系式 6D 可供性图和约束轨迹生成实现关节物体操作。
教育经历
2024.09 - 至今 - 上海交通大学巴黎卓越工程师学院，电子信息硕士
2020.09 - 2024.06 - 上海交通大学巴黎卓越工程师学院，法语专业，辅修信息工程（IE）
实习经历
2026.03 - 2026.07 - 智元机器人：手机包装任务强化学习后训练与轮足机器人全身控制
2024.06 - 2024.09 - 阿里云：基于 ViNT、NoMaD 的视觉导航微调，解决训练数据直行偏置问题
2022.06 - 2022.08 - 阿里云产品实习：负责杭州亚运会首款零碳特许商品吉祥物的产品碳足迹认证，并辅助中国美术学院低碳平台
技能
编程：Python、PyTorch、JAX、C++、Java、MATLAB · 方向：VLA、VLN、智能体、世界模型、真机 RL