A robot can learn to move an object across a workbench and still struggle when the bench is higher, the object is a different size, or tools get in the way.
Robotensor is a Physical AI company building the data and software infrastructure to help robots adapt to those changes.
Its approach, known as in-context robot learning, gives a trained model a demonstration to follow while keeping its weights frozen. The model uses the example to decide how to act in the scene in front of it.

Robotensor develops this approach through Attune (SN58), its robotics subnet on Bittensor. The company collects demonstrations, uses public competitions to measure how well models act on them, and gives developers a way to buy the resulting training data.
That combination gives Robotensor a way to develop adaptive robotics models and put the data behind them to work outside the competition.
How Attune Trains and Tests Robot Models
Attune runs two programs. Vector measures how well robot models adapt to unfamiliar scenes, while Horizon collects the demonstrations those models learn from.
1. Vector Tests Adaptation: In each public duel, two models receive one expert demonstration of a task that requires two robotic arms. Both face the same randomized scene, generated from a blockchain seed after submissions are committed. Models share a fixed architecture and keep their weights frozen during testing.

2. Horizon Collects Demonstrations: Contributors can use a smartphone to control a simulated robot arm while watching on a computer, or use a simple head mount to film everyday tasks from their perspective.

Accepted submissions earn points that determine each contributor’s weekly share of a public vault funded by part of Attune’s miner emissions. A physical robot is unnecessary for either contribution method.
Horizon gives model builders both simulated robot movements and footage of people doing everyday work. Vector gives them a shared test for whether the models trained on those demonstrations can handle changes in objects and surroundings.
Gamma Opens Attune’s Data to Buyers
Attune was the first subnet to go live with Gamma, Bittensor’s on-chain usage-credit system. Researchers, companies, and developers inside and outside Bittensor can spend Gamma credits to access Horizon’s training data.
The network records usage on-chain, giving Attune a way to earn revenue alongside the emissions that fund participation.

Robotensor plans to extend Gamma payments to model access and a playground where developers can bring their own models.

How Attune Compares With Other Bittensor Robotics Subnets
OpenRoboto (SN80) is the closest comparison because both subnets use competitions to improve robot models. OpenRoboto combines benchmark tournaments and physical robot testing with a pipeline for first-person workplace footage through Shift. Attune uses reproducible simulation duels to measure whether a model can use one demonstration to perform a task in a different scene.
THOR (SN112) focuses on preparing training data for robotics labs by annotating footage and scoring the work against client requirements. Swarm (SN124) focuses on autonomous drone flight, using simulation and physical hardware to test how well flight policies transfer.
Attune pairs phone-based data collection with a competition focused on adaptation in tasks that require two robotic arms.
The Next Test for Robotensor
Robotensor is building a way for robotics teams to collect, test, and reuse demonstrations. Horizon broadens the range of examples available for training by letting people contribute with a smartphone. Gamma gives teams outside Attune a way to use that data in their own robotics projects.
Future Vector stages will test whether the same demonstration can guide robots with different designs and whether a robot can learn a task by watching a person do it. If those tests succeed, teams could record a task once and reuse that demonstration across the different machines they already operate.
Enjoyed this article? Join our newsletter
Get the latest TAO & Bittensor news straight to your inbox.
We respect your privacy. Unsubscribe anytime.
Enjoyed this article?
Join our newsletter
Get the latest TAO & Bittensor news straight to your inbox — every morning before markets open.





No comments yet — be the first.