Sunqi Fan

CS Ph.D. Student @Tsinghua_Uni, Long-time Intern @TencentHunyuan @TencentAI_News @TencentGlobal

[rpzhou @rpzhou]

Delighted to see Simple-WAM top the Hugging Face "Daily Papers" list. Although rankings fluctuate in real-time, we would like to take this moment to briefly introduce the latest work from our LeapLab team.

[Sunqi Fan @Sunqi_Fan]

@rpzhou solid works!!🎊🎊

[Sunqi Fan @Sunqi_Fan]

@zhengyang_42 congratulations! 🎊🎊

[Sunqi Fan @Sunqi_Fan]

See our paper, results and episodes on https://t.co/PvuW5NmT6t ! #embodyai #Robotic #gpt #robodawn #harness #astra

[Sunqi Fan @Sunqi_Fan]

Excellent work by @MengHaoGuo1 and @JiapengZha233 !

[Sunqi Fan @Sunqi_Fan]

📈 In-context learning: one demo lifts RoboTwin with GPT-6-Astra 2.0 53.2% → 73.6%, RoboDojo 35.67% → 47.17%
🔁 Test-time scaling: 31.2% → 47.2% with more commands
🔓 710 episodes and sessions of RoboTwin & Robodojo open as full replays

[Jiapeng Zhang @JiapengZha233]

Introducing our work: RoboDawn 🌅 A VLM controls robots with zero robot training.

📈 In-context learning: one demo lifts RoboTwin with GPT-6-Astra 2.0 53.2% → 73.6%, RoboDojo 35.67% → 47.17%
🔁 Test-time scaling: 31.2% → 47.2% with more commands
🔓 710 episodes and sessions of RoboTwin & Robodojo open as full replays

See our paper, results and episodes on https://t.co/HbIujXuRzR !

#embodyai #Robotic #gpt #robodawn #harness #astra

[Sunqi Fan @Sunqi_Fan]

目前比较感兴趣的方向包括：
LLM / MLLM
Generative AI
Robotics
其他与多模态和大模型相关的研发岗位

如果腾讯内部有团队正在招聘实习生 / 有合适的转岗机会，或者你恰好了解相关团队，非常欢迎和我联系，也非常感谢任何 referral、团队推荐或信息分享。

[Sunqi Fan @Sunqi_Fan]

过去一段时间，我在腾讯混元多模态视觉中心担任实习生，主要参与多模态 / 视觉方向的相关工作。近期由于团队组织与业务调整，我也开始主动探索腾讯内部新的团队和转岗机会。

#Tencent #LLM #MultimodalAI #ComputerVision #VLM #GenerativeAI #AI #MachineLearning #OpenToOpportunities

[Sunqi Fan @Sunqi_Fan]

这是我最新的简历：
https://t.co/8thyHK6Mvw
这是我的主页：
https://t.co/m11ZdktspZ

[Sunqi Fan @Sunqi_Fan]

[Thread 5] Our goal is to help GUI agents move from simply understanding videos to learning transferable workflows from tutorials.

Code and data are fully open-sourced!

#GUIAgent #VideoUnderstanding #MultimodalAI

[Sunqi Fan @Sunqi_Fan]

From VideoQA to video-guided agentic tasks: our work has been accepted to #ECCV2026! Code and data are now open-source.

Project Page: https://t.co/3k5GBo5Azc
Code: https://t.co/370mRIqAc2
Arxiv: https://t.co/w4hMdrOdZI
Huggingface Demo: https://t.co/SA1QJZx2ki

[Sunqi Fan @Sunqi_Fan]

[Thread 4] Since not every video frame is equally useful, we propose a training-free keyframe selection method (TAKSER, a task-driven and scene-aware keyframe search algorithm).

It focuses on action-critical moments and improves both VideoQA and video-guided agentic tasks.

[Sunqi Fan @Sunqi_Fan]

[Thread 3] Beyond answering questions about a video, agents must:

🎬 Understand demonstrated actions
🧠 Extract reusable procedures
🖱️ Complete new tasks in unseen interfaces

[Sunqi Fan @Sunqi_Fan]

[Thread 2] Can MLLM-based GUI agents learn procedural knowledge from video tutorials—and transfer it to new GUI tasks? 🧵

We introduce a GUI agent benchmark paired with video tutorials, named VG-GUI-Bench (Video-Guided GUI Agent Benchmark).

[Sunqi Fan @Sunqi_Fan]

@ZhizhouSha Thank you bro🫡🥰

[Sunqi Fan @Sunqi_Fan]

@lupantech @UCLAComSci @kaiwei_chang @uclanlp Congrats!!!