Today we’re launching OpenRoboto Shift, opening a new chapter for OpenRoboto. Shift is a decentralized network for collecting egocentric robotics data: first-person video of real people doing real work. Only Possible on Bittensor. Explore Shift →

pre-trained DYNA-2, a robot-learning model, on first-person human video. Increasing that data from 1,000 to 1,000,000 hours raised its average score across 14 real-robot tasks from 20% to 53%. It is a scaling law. The answer is more hours.

Figure (@Figure_robot) launched Figure Index, an app that pays people to record physical tasks for robot training. At launch, it had over 44,000 weekly active contributors and downloads across 108 countries. Figure also committed more than $1B to data and compute over the following 12 months. We built OpenRoboto Shift as a decentralized counterpart on Bittensor.

Shift connects robotics companies that need training data with operators who organize collection in real workplaces. Operators can apply to collect with OR-S1 or OR-S2, our head-mounted stereo capture devices. Each recording stays traceable to its source. We’re opening a new data track on Subnet 80 and allocating part of the subnet’s emissions to reward accepted recording hours.

Buyers get exclusive access to the data they purchase for an agreed period. Once that period ends, we make it available through the Open Data Pool, where researchers and OpenRoboto miners can use it to train models.

Shift is open to operators now. If you can organize data collection in factories, kitchens or other workplaces, we want to hear from you. Apply →

Expanding and progressing. Things we love to see 🦾

👀

@ShizzyUnchained Is $ROBOTO = base:0x6f63d869011f95274498023b4abfc00b30c34378 ? What’s the projected ROI if we buy the device? Will we be able to test the device at @ExploitSummit ?

you guys move fast, impressive!

Read about our visions here: