Shreyas Gite (@shreyasgite) 's Twitter Profile
Shreyas Gite

@shreyasgite

Robots, syn data, & animal tech at tuul.ai. Prev: founded self-driving at Kopernikus (acq. @Ford), jet engines @RollsRoyce.

ID: 78009797

linkhttps://www.tuul.ai calendar_today28-09-2009 13:12:13

891 Tweet

604 Followers

183 Following

Shreyas Gite (@shreyasgite) 's Twitter Profile Photo

For my constrained case of Toy sorting; I found affordances pretty handy to easily generalise -> you could just talk the robot. Hence went down a little rabbit hole of affordances, and here is quick summary: 1. Raw RGB (+ proprioception) Where the inductive bias lives:

Pablo Vela (@pablovelagomez1) 's Twitter Profile Photo

With that slight labeling detour, I'm back to ego/exo data. Really pushing things on the Rerun viewer with 8-1080p exo videos + 4-VGA ego videos. Will have more on Friday or next Monday, but I'm trying to consolidate all of the exocentric + egocentric datasets into a single

Paszea (@paszea) 's Twitter Profile Photo

The ground is shifting for robotics.🤖 And LeRobot made it easy for all to experience this shift firsthand.✨ I wrote my experience here bit.ly/4jV2snq (w/ videos, datasets, learnings!) The gist:  a total time-suck but the fun & learning 💯 worth it.

Shreyas Gite (@shreyasgite) 's Twitter Profile Photo

Then there is meta-learning. The more diverse data we have, the better the model gets at learning representation. Human equivalent example being how Federer tried a dozen sports (skiing, wrestling, swimming, skate-boarding, basketball, handball, table-tennis, badminton, soccer,

Ted Xiao (@xiao_ted) 's Twitter Profile Photo

Big things come in small and fast packages: announcing the Gemini Robotics On-Device VLA model!🤖 Our new model is optimized for local inference, while showcasing many properties Gemini Robotics excelled at: strong generalization, instruction following, and fast adaptation.

Big things come in small and fast packages: announcing the Gemini Robotics On-Device VLA model!🤖 

Our new model is optimized for local inference, while showcasing many properties Gemini Robotics excelled at: strong generalization, instruction following, and fast adaptation.
Shreyas Gite (@shreyasgite) 's Twitter Profile Photo

Data vs Priors in Robotics, this was fun! Best part was house-mate of Rich Sutton popping up at 44 mins:D Most agreed with the data will solve robotics part, just framed it differently:D The only counter argument for state space models was energy/climate and likely not being able

Patrick Hsu (@pdhsu) 's Twitter Profile Photo

Delighted to announce Arc Institute's Virtual Cell Challenge - a recurring, open, community-driven challenge to benchmark cellular foundation models See our announcement in Cell below, with prizes up to $100,000, sponsored by @Nvidia 10x Genomics Ultima Genomics!

Delighted to announce <a href="/arcinstitute/">Arc Institute</a>'s Virtual Cell Challenge - a recurring, open, community-driven challenge to benchmark cellular foundation models

See our announcement in <a href="/CellCellPress/">Cell</a> below, with prizes up to $100,000, sponsored by @Nvidia <a href="/10xGenomics/">10x Genomics</a> <a href="/UltimaGenomics/">Ultima Genomics</a>!