Twitxy Saved

Ajay Subramanian

@ajaysub110

Staff Research Scientist @BostonDynamics | RL for dexterous manipulation 馃馃
Cambridge, MA ajaysubramanian.com Joined March 2018
392 Following 1K Followers 444 Posts

Latest posts 5

Ajay Subramanian @ajaysub110
Wow! Scaling-up pre-training data in VLAs enables learning from egocentric human data. LLM-like patterns in robotics!
Simar Kareer @simar_kareer Dec 17, 2025
I did similar experiments at Georgia Tech, but didn鈥檛 see this level of transfer. The missing piece was that I lacked a strong enough foundation of robot data. More robot data in pretraining -> more transfer from human data
View quoted post on X