Dechen Gao (@DechenGao) profile picture

🤖 Robot Learning | Research Scientist Intern @ Meta Reality Labs, Ex-Amazon, TikTok, Meta | CS PhD @ UC Davis

Davis, CA
Joined February 2023
Photo by Dechen Gao (@DechenGao)Photo by Dechen Gao (@DechenGao)Photo by Dechen Gao (@DechenGao)Photo by Dechen Gao (@DechenGao)Photo by Dechen Gao (@DechenGao)Photo by Dechen Gao (@DechenGao)
🦾 We're generating dexterous robot demonstrations from human data at scale 🌟 Check out our work at Meta Reality Labs Research 🔥 Generative Neural Retargeting for Human-to-Robot Dexterous Manipulation 🌐 Website: yy-gx.github.io/GNR/ 📚 Report: arxiv.org/abs/2610.12440
7
27
6
270
18,971
🤖 We generate dexterous demonstrations from human data at an unprecedented scale, retargeting 223k human demonstrations, and solve tasks requiring <1mm precision over half a minute. Congrats to our amazing team! @YYang9923 @frankzydou @Outsider_pc @babbatem @ewhitmire
5
1,124
Dechen Gao retweeted
Excited to share that Personalized RewardBench has been accepted to COLM2026! 🎉🍻 Reward models know what's good — but do they know what you like? Our answer: not really. By controlling for general quality, we force models to compete strictly on personal preferences. Come see if your RM can actually read the room 👀 @COLM_conf #COLM2026
3
4
38
3,237
Dechen Gao retweeted
Can SOTA reward models truly handle "pluralistic alignment"? Spoiler: Not really. Introducing Personalized RewardBench! By controlling for general quality, we force models to compete strictly on user-specific rubrics. See if your RM can actually read the room! 😎👇
1
4
7
213
🤖Check out our ICLR 2026 paper: Vision-to-Action Flow Matching Policy (VITA). ⭐️Instead of sampling from Gaussian, VITA directly flows from latent images → latent actions, enabling efficient and high-precision robot policies. Website: ucd-dare.github.io/VITA/
3
1
189
Codes, datasets, and more demos: ucd-dare.github.io/VITA/
38
VITA is the first MLP-only visuomotor policy that succeeds on complex real-world bimanual tasks, thanks to the closely aligned visual & action manifolds, whereas MLP-based FM struggles due to the need to transport from an unstructured Gaussian prior to a structured action space
34
Surprisingly, visual latents in VITA manifest action semantics. The latent image can decoded into a smooth trajectory, and progressively refined by the ODE process. VITA closely aligns the visual and action latent manifolds.
118
Dechen Gao retweeted
My summer research at Tencent is finally public 🤯 We released a generative reward model with critical‑thinking abilities, built for RL post‑training of LLMs in QA with supporting docs. 📜 arxiv.org/abs/2509.25409 🖥️ github.com/Martin-qyma/TRM 🤗 huggingface.co/QiyaoMa/TRM
1
6
316
Humans actively move their heads to find best perspectives to solve tasks. Can robot do the same? Check out our latest work!
In our joint work with @ian__chuang , @andrewleecw, and @DechenGao , we explored how mimicking human behavior—specifically, the natural use of head and body movements to find optimal viewpoints—can improve robotic manipulation. soltanilara.github.io/av-alo… arxiv.org/abs/2409.17435
2
138