Roadmap

THOR begins with verified robotics data. The longer-term direction is to build on that foundation with open models, hardware and physical deployment. These are planned stages, with scope shaped by customer needs and what earlier work demonstrates.

1. Stage 1
   Data

   Useful, verified data

   Start with a private annotation beta, then open annotation to permissionless participation.

2. Stage 2
   Models

   Open model development

   Build on data and evaluation methods to train and test models for defined tasks.

3. Stage 3
   Hardware

   Physical systems

   Develop capture tools and test systems in the field, returning new experience to the data layer.

Progress follows evidence and operating readiness, without fixed deadlines.

Data: the starting point

THOR starts with a private commercial annotation pilot: the beta. Qualified, whitelisted professional annotation firms work against buyer-defined specifications, acceptance criteria and commercial delivery requirements.

After beta, annotation will open to permissionless participation. This transition belongs to the Data stage. Task-specific requirements and customer data permissions will still apply. Participation explains the access path.

The beta establishes the production and review workflow and tests verification and delivery costs. Later data work could extend into complete dataset commissions around particular research tasks and target hardware. Agent-directed requests and crypto settlement are longer-term ambitions within that market.

Progress should be demonstrated through accepted deliveries, consistent review decisions, manageable correction effort and sustainable production and review costs. The pilot should also show that source permissions and delivery records remain traceable. The beta will provide evidence against these readiness questions.

Models: build on the foundation

Data and evaluation methods can support open robotics model development. Research and simulation can help assess whether those models improve at the tasks that matter. The scope of this stage remains to be developed.

Hardware: connect learning to physical systems

Capture equipment and interfaces can expand the supply of compatible physical experience. Bounded deployment projects can test the combined system under operating conditions and return failures as new data or evaluation requirements.