Alibaba · Qwen
Qwen-VLA
A generalist policy across tasks, environments, and robot embodiments.
Hello! I’m Xintong
I build robot policies that follow fine-grained instructions.
I am a first-year PhD student at XLANG Lab, The University of Hong Kong, advised by Prof. Tao Yu. My research focuses on embodied AI, especially vision-language-action models.
I received my B.S. in Computer Science and Technology from Zhejiang University, where I was advised by Prof. Zhou Zhao.
As a research intern with Alibaba’s Qwen team, I contributed to Qwen-VLA and built a verifiable coding-data pipeline for Qwen3.8 and Qwen3.9. I am also a first author of FineVLA, a framework for aligning robot policies with fine-grained instructions.
Highlight
XLANG Lab && Qwen
FineVLA
Steerable policies trained with fine-grained instruction alignment.
Latest updates
News
Research output
Publications
* Equal contribution
Full list on Scholar ↗
GitHub stars · Semantic Scholar citations
906 Stars
64 Citations
2026NeurIPSFineVLA
FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies
An open framework for fine-grained VLA supervision: human-verified trajectories, a fact-level benchmark, a robotics annotator, and a steerable dual-arm policy.
2026Technical ReportQwen-VLA
Qwen-VLA: Unifying Vision-Language-Action Modeling across Tasks, Environments, and Robot Embodiments
A unified embodied model that extends Qwen’s vision-language stack to action and trajectory generation across manipulation, navigation, and robot embodiments.
2025NeurIPSMRSAudio
MRSAudio: A Large-Scale Multimodal Recorded Spatial Audio Dataset with Refined Annotations
A 500-hour multimodal spatial-audio dataset with ambisonic audio, synchronized video, motion, and fine-grained annotations, plus a benchmark for five spatial audio tasks.
2025NeurIPSTree of Preferences
Tree of Preferences for Diversified Recommendation
Uses an agent to complete user context and improve recommendation diversity, addressing filter bubbles in collaborative filtering.
Selected work
Projects
2026Alibaba · QwenCoding data
Verifiable Coding Data and Rollout Pipeline for Qwen3.8 / Qwen3.9
Turns real IDE assistant logs into reproducible coding tasks: mined multi-turn cases, paired buggy and resolved repository states, tests that fail before the fix and pass after it, and containerized rollouts labeled by hidden tests.
Research journey
Internships
01
Alibaba · Qwen
Qwen-VL Team
Research Intern · Advisor: Shuai Bai
VLA & verifiable coding agentAcademic background
Education
Recognition
Honors & Awards
National Scholarship · Top 1%
First-Class Scholarship · Zhejiang University
Top 3%Second Prize · IEEE ASRU AudioMos Challenge
Shangde Scholar · Zhejiang Province
Single recipientCampus Star · Zhejiang University, Computer Science
Top 10 studentsNational Scholarship · Top 1%
First-Class Scholarship · Zhejiang University
Top 3%
Contact
Let’s make robots move on.
One step forward for robots that see, decide, and act.