Building robot intelligence requires high-quality robot data. But far too many tools to collect data are closed and custom-built. Open Teach is fully open-sourced and is: - Calibration-free - Supports multiple arms, hands, & mobile manipulators - Costs just $500

A few reasons why Open Teach works so well: 1. It is focused on low latency & high-frequency visual feedback. 2. Native passthrough from Quest 3 gives us exceptional visual clarity for 3D task perception. Here is a teleop of making a sandwich.

Open Teach is really versatile! We tried it on 38 tasks spanning 5 real robots and 2 simulated platforms!

And that's not all! Data from Open Teach is compatible with data-driven learning. On 10 tasks across platforms, we achieves a 87% success rate.

Open Teach is fully open sourced! Website: Paper: Code: By @AadhithyaIyer, Bobby Peng,  @wxdyl0915, @haldar_siddhant, @irmakkguzey, @soumithchintala

This is awesome. Congrats, and keep on pushing. I'm excited about the future of robots. This technology has so many use cases that will change our lives.

What happens when someone uses it to hurt other people? Will it be treated as an accident? Do we have any law that governs these types of cases?

What do you think of Figure 1, just released with openAI??

@ylecun Moravec’s Paradox 😬 great work though

Does this use imitation learning? If I teach it to make a tuna sandwich in my mom's kitchen, can it then make a chicken sandwich in my dad's kitchen, or do I have to train it again for each sandwich, for each kitchen, etc?

@ylecun @Jesusgomez