Tired of collecting robot data? Check out our new autonomous data collection and policy improvement system from Paul Zhou and Pranav Atreya!

Can VLMs enable robots to autonomously improve? In our new work we ran a fleet of robot arms to collect autonomous data with VLM-proposed tasks and showed that robots can keep getting better as they are deployed, without supervision: auto-improvement.github.io 🧵👇

Wow, this humanoid 🤖 **and** it's 16K$ price tag are a game changer!

We don't have much evidence for that statement. Look at the famous videos from the 2007 Stanford PR1, they already showed impressive skills in unstructured environments, but with teleop. Progress towards generalist robots has clearly been driven by better robot learning software.

Huge shoutout to the amazing Ria Doshi and Homer Walke for leading this project and to collaborators Sudeep Dasari Sergey Levine We release the paper, code and weights at crossformer-model.github.io For more details about our approach and results, please check out Ria's thread below!

Can one transformer-based policy fly drones, manipulate objects, walk, and drive? Ria Doshi, Homer Walke, Oier Mees, Sudeep Dasari show in their new papers that CrossFormer can do all of these things, with flexible per-embodiment action heads: crossformer-model.github.io

Back by popular the demand, the 3rd Workshop on Language and Robot Learning will take place at #CoRL2024 in Munich! Submit your papers and come learn what's next for Foundation Models and robotics!

Can generalist policies quickly capture a new task w/ just a few demos? In PALO, we imitate a task the way people do: not by copying actions, but inferring the steps and repeating them with a generalist policy. CoRL paper w Vivek Myers Bill Zheng Oier Mees Kuan Fang 👇🧵

If you are looking for a PhD position in robot learning consider applying to work with Prof. Shah!

Honored to see our recent work on cross-embodiment learning featured on IEEE Spectrum! I truly believe generalist policies need to be cross-embodied.