←
THOR Group36 posts
THOR Group avatar

THOR Group

@THORMachines

Home of Open Source Robotics

◷ Joined February 2025▧ 15 Photos and videos
THOR Group avatar
THOR Group@THORMachines·Oct 14

The Age of Machines Should Belong to All of Us.

1:38
THOR Group avatar
THOR Group@THORMachines·4h

NVIDIA is teaching robots physical work by showing them video of people doing it. Its new model, GR00T 2, was pretrained on about 17 hours of human footage for every hour of robot teleoperation data. Teleoperation requires a real robot and an operator. Video of people doesn’t, so it scales faster. That footage is increasingly open. In November alone, @NVIDIArobotics, @GroundedSI and @hubdotxyz released about 120,000 hours under licences that allow commercial use. To find specific actions in those hours, labs need labels: what happens, and when it starts and ends. These releases do ship with labels, but each follows its publisher’s own definitions and review process. GroundedSI says people reviewed captions covering 1,000 of its 10,000 hours. Hub says its labels have not been checked by people. Our read: as more labs train robots on video of people, more of the footage they use for specific tasks will need labelling and checking to their own definitions. Where existing labels don’t meet a lab’s needs, THOR helps get the footage labelled that way and verified against its criteria. THOR Signals · Datasets

Post attachment
THOR Group avatar
THOR Group@THORMachines·Nov 12

$20K. THOR’s first customer deposit landed onchain before work began. This deposit was matched within our whitelisted miner set for annotation work. Miners are paid as work passes verification. Buyers pay only for accepted units.

Post attachment
THOR Group avatar
THOR Group@THORMachines·Nov 10

A data network should start with customer demand. THOR’s first jobs come from signed robotics labelling contracts. Customers fund each job upfront into an onchain deposit contract and pay only for accepted units. A share of those payments funds SN112 alpha buybacks.

Post attachment
THOR Group avatar
THOR Group@THORMachines·Nov 6

16 seconds of someone unloading a dishwasher. A THOR miner turns that footage into a structured record: eight actions, each with a start and end time. At scale, labels like these become training and evaluation data for robots. Action labelling is THOR’s first task type. Every THOR task is built to the customer’s brief, and work is checked before delivery.

0:22
THOR Group avatar
THOR Group@THORMachines·Nov 4

“Model training is not hard. But I need a lot of data… to train my robots.” We hear this from robotics labs constantly. Software AI learned from the internet. Robots must learn from the real world, where usable data is still scarce. That’s why THOR starts with data.

1:34
THOR Group avatar
THOR Group@THORMachines·Nov 3

October was THOR’s first month live on Bittensor SN112. • First robotics labelling contracts signed • 3 annotation firms onboarded • Invite-only orderbook • talorobotics.ai launched November is about delivery: signed contracts → accepted work → subnet revenue.

Post attachment
THOR Group avatar
THOR Group@THORMachines·Oct 30

Sources: Figure, Introducing Index (25 Aug): figure.ai/news/introducing-i… Mecka Series B (7 Oct): techcrunch.com/2026/10/07/ro… Maxinsights seed (22 Oct): roboticsandautomationnews.co… Pantheon, Argus audit (1 Oct): pantheon.inc/research/argus

THOR Group avatar
THOR Group@THORMachines·Oct 30

THOR Signals · Capital @Figure_robot tried to buy robot training data. Vendors couldn’t hit the throughput, diversity or quality it needed. So it built its own pipeline, and in August committed $1B+ to data and compute over 12 months. Meanwhile, data suppliers are raising: → @MeckaAI: $60M Series B led by Sequoia, with Nvidia and M12 joining. → @MaxinsightsAI: $30M seed led by Peak XV. 2M+ hours of human activity video captured, with 1.5M rated “model-ready” under its internal scoring system. As more capital flows into collection, more data needs verifying, and that’s where things slip. Pantheon audited 3,546 episodes across nine open robotics datasets and found 27% had at least one medium or high severity issue. That's open data, not these suppliers’, but it shows how easily problems get through. Figure’s answer: keep filtering, fraud review and deduplication in-house. Our view: an open market for robotics data needs verification buyers trust enough to cut back their own checking. THOR’s pilot is testing whether independent verification can match buyers’ own quality calls, and cut the time they spend checking.

Post attachment
THOR Group avatar
THOR Group@THORMachines·Oct 30

Per Scale’s paper, FineART’s subtask training raised success from 32% to 100% on one spatial task. Building that dataset took three teams: one to collect, one to annotate, one to verify. Many labs can’t staff three teams. They need open infrastructure that can do that work.

In collaboration with @HuggingFace, we’ve released the largest open robot dataset, alongside their first open System 1-System 2 model that can reason over step-by-step actions. FineART is built with 40K+ demonstrations across 151 tasks and 500K subtask labels. Together, the dataset and model were designed for each other, and both are available on @LeRobotHF where robotics developers work today.

2:43
THOR Group avatar
THOR Group@THORMachines·Oct 28

“The main lesson of thirty-five years of AI research is that the hard problems are easy and the easy problems are hard.” Steven Pinker, 1994 Robots still struggle with what a toddler does without thinking. THOR is building the data network to help them learn.

THOR Group avatar
THOR Group@THORMachines·Oct 27

If everyone has access to better AI models, what do contributors compete on? How well they use them. Models like GPT-6 Astra and NVIDIA’s Flora are getting much better at labelling robotics video. NVIDIA estimates inference compute costs below $1 per hour of footage for Flora. But customer-ready data has to meet timing, description and quality requirements. Verification and rework carry costs too. On Bittensor Subnet 112, THOR is designed to tie rewards to verified output, giving contributors an incentive to improve their models, tools and workflows. A better workflow raises the bar for everyone. The test for THOR is whether that competition delivers reliable robotics training data at a lower total cost. Better AI expands what’s possible. Competition makes it count.

Post attachment