Robotics Ecosystem
We believe robotic learning is starting to mature as an ecosystem. Companies are beginning to gain an advantage by focusing on individual components rather than building everything themselves. The core areas we see are hardware, models, data, and applications.
We believe much of the value for deployment comes from deep knowledge of an application. However, people closest to these applications often have limited visibility into the state of the art in robot learning, and where it fits in with classical robotics. We want to give them visibility into model performance, and best practices, giving them confidence to try out robotics where it is right.
Where We Fit In
Playground — We opened up our arms to allow people to play with them and gain an intuition for how well they actually perform.
Benchmark — We are in the process of opening up a benchmark on a few tasks. This will include tasks that are interactive and practical. Releasing the first version soon.
Currently, benchmarks and demos do a poor job of showing what actually works. Teams will often build their own stack and run multiple experiments to find out.
We hope to give teams a more practical intuition for what actually works, so those looking to deploy models can select tasks that are a good fit for them and follow common practices to limit the amount of data and teleoperation needed.
Research ideas
We’re really excited about these ideas, and we’re looking for people to work on them with us. Help turn experiments with real robots into more deployments.
Code As Policy
Let an LLM reason through the setup, then use the code it produces for fast robot execution.
Distillation from Astra
Train faster robot policies from Astra’s demonstrations and use its reasoning to generate corrections when they fail.
Is the sim-to-real gap a thing of the past?
Explore whether LLM reasoning can focus on what matters for a task, so simulation becomes a better guide to real-world performance.
The $100 Ducky Bounty: stack three blocks and place a ducky on top using the playground. The first person to do it wins $100.
Explore the research ideas