Robot intelligence built in the open. Bittensor subnet 80.
RoboVerse
- Tweets 156
- Following 128
- Followers 3,566
- Likes 485
Pinned Tweet
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 → shift.openroboto.ai/
15
47
239
49,450
The world is full of human experience that robots could learn from. The missing piece is a way to coordinate and reward its capture.
That’s what we’re building with OpenRoboto Shift.
A global learning network for open robot intelligence.
shift.openroboto.ai
To make robots work, the key bottleneck is DATA.
@openroboto is solving that in the Bittensor world.
@Figure_robot has been crowdsourcing data in the humanoid world.
2
5
45
1,253
OpenRoboto retweeted
To make robots work, the key bottleneck is DATA.
@openroboto is solving that in the Bittensor world.
@Figure_robot has been crowdsourcing data in the humanoid world.
OpenAI shut down its robotics team in 2021 explicitly because of the data problem, then restarted it later. having the best LLM didn't solve that then.
everyone assuming a frontier VLM solves robotics runs into the exact same wall:
action data is the actual bottleneck. the internet has no torques, contact forces, proprioception, or failure-recovery trajectories. pretraining compounds for the "what," while the "how" has to be physically collected, costs per hour, and doesn't scale like tokens.
embodiment fragmentation breaks the data pool further. physical data is partly tied to specific hands, kinematics, and sensors. cross-embodiment transfer helps, but it isn't free.
evals happen in the real world and take days, not minutes. that blunts the core advantage of frontier AI labs, which is fast scaled iteration.
the long tail is operational. reaching 99.9% reliability in messy sites means dealing with hardware, safety, maintenance, and customer ops. that is a low-margin, atoms-heavy business, a poor fit for a software-margin org.
the model backbone is commoditizing anyway. open VLMs are good-enough starting points. Physical Intelligence, Skild, and others built competitive policies without owning a frontier LLM.
1
4
37
3,786
@openroboto pushed SN80 deeper into robotics data
OpenRoboto launched Shift, a decentralized network designed to collect first-person footage of real-world tasks and turn it into training data for robots.
That gives SN80 a way to source the physical-world data needed to train and improve robotics systems, rather than relying purely on synthetic or benchmark data.
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 → shift.openroboto.ai/
1
15
349
Honored to join with the Verona team once again to co-host another event during KBW in Korea! 🇰🇷
luma.com/z20755mp
See you soon!
On Monday, Verona makes its biggest announcement of the year, and the next afternoon our team will walk you through it in person.
We're hosting on September 29 at Spark Plus Seolleung, with @GoKiteAI, @AethirCloud, @openroboto, and @De1_ai as co-hosts.
1
4
45
1,710
Introducing the Open Axis Benchmark — a living benchmark engine for honest evaluation of robot manipulation models built on @axisrobotics' data engine and co-developerd by OpenRoboto.
Submit a model. Prove skills that transfer.
Live on openroboto.ai for model evaluation
Introducing the Open Axis Benchmark, a living benchmark engine for honest evaluation of robot manipulation models, built together with @openroboto.
Robotic models evolve faster every month, while most benchmarks stay frozen. Models overfit to fixed task sets, scores stop reflecting real generalization, and that distorted signal misleads and holds back model evolution.
Open Axis Benchmark isn't just larger. It lets evaluation itself keep pace with the models: each round locks a fresh task set drawn from the growing Axis Library, so models prove they can generalize instead of memorizing the test.
Axis provides the data engine behind it, continuously producing tasks at the scale and diversity needed to genuinely test generalization. With Open Axis Benchmark, the same compounding loop that trains robots also powers how they are evaluated.
Submit your model: openroboto.ai
Docs: openroboto.ai/#/docs
See how it works ⬇️
6
15
97
4,413
Tasks rotate out when they stop separating models — when everyone clears them, or nobody does. A task like that no longer changes anyone's result, so it leaves.
Replacements are drawn from the Axis library: 6,000+ tasks, growing as Axis collects more. Each round's set is drawn, frozen, and published with a version number before submissions open.
That makes overfitting extremely difficult here. You can overfit one round. You can't overfit the library — a model that clears 6,000+ tasks across every category has, in fact, generalized.
1
11
230
Open Axis Benchmark is live on the simulation track.
Check out our docs before you submit
openroboto.ai
3
124
Can't wait to see the results!
We’re excited to announce a pilot with @OpenRoboto to support the growth of OpenRoboto Shift.
Shift is building a decentralized network to collect real-world training data for robotics. Leadpoet will help identify and engage the enterprises that can power it, from factories and warehouses to kitchens and hotels.
The partnership combines high-intent lead generation with targeted outreach to bring these businesses onto Shift as enterprise miners.
SN71 × SN80. Two Bittensor subnets connecting their capabilities to drive global enterprise adoption.
2
10
85
2,433
We need enterprise miners with real work sites. @LeadpoetAI finds the companies that have them.
OpenRoboto (SN80) and Leadpoet (SN71) are partnering to bring them onto OpenRoboto Shift — starting with a pilot package covering high-intent leads and outreach.
4
10
109
1,851
We just announced OpenRoboto Shift — a decentralized network collecting the egocentric data robots learn from.
That data only exists where physical work already happens: factories, warehouses, kitchens, hotels. Those are the leads we need — and that's exactly what Leadpoet is designed for.
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 → shift.openroboto.ai/
1
21
1,069
Two incentive networks, one loop. Only Possible on Bittensor.
Run a site where real work happens? → shift.openroboto.ai
10
169
OpenRoboto retweeted
We are super stoked to support @openroboto's mission to build a decentralized robotics data market place. GI devices + Openroboto collector network will create a massive amount of Physical AI training dataset.
With Shift now live, we’re unveiling OR-S1, our stereo egocentric capture device for robot manipulation data.
We’ve partnered with @gi_labs to bring it to market.
Pre-order OR-S1 → shift.openroboto.ai/
2
4
30
2,628
🤝🤝🤝
We are super stoked to support @openroboto's mission to build a decentralized robotics data market place. GI devices + Openroboto collector network will a massive amount of Physical AI training dataset.
14
1,015
With Shift now live, we’re unveiling OR-S1, our stereo egocentric capture device for robot manipulation data.
We’ve partnered with @gi_labs to bring it to market.
Pre-order OR-S1 → shift.openroboto.ai/
15
20
140
25,250
@DynaRobotics pre-trained DYNA-2 on more than one million hours of egocentric human video, without robot action data during pre-training.
Its experiments demonstrated a human-to-robot scaling law: more human video improved predictions on unseen robot data. Those gains carried through to real robots after post-training with robot demonstrations.
First-person video lets us scale collection across people and workplaces, without a robot at every recording site.
Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws:
• world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours,
• this human data scaling law implied a scaling law on never seen robot data,
• both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge
🧵
1
13
646
The first OR-S1 units will go into the field through OpenRoboto Shift.
We’re building collection around people doing their actual jobs, capturing the physical tasks we want robots to learn.
If you’re collecting manipulation data at scale, or planning to start, get in touch.
Pre-order OR-S1 → shift.openroboto.ai/
5
206