Documentation / Start here
04 / 17
What openness means
Shared foundations that people can inspect, adapt and build on.
On this page
Much of software AI's power sits with a small number of platforms controlling models, compute and data. Robotics risks similar concentration as companies build proprietary fleets, data loops and models that outsiders cannot inspect or reuse.
Proprietary platforms
OpenAI
Anthropic
Google
Data + models + compute
A few platforms control the full stack
Open robotics infrastructure
Global research community- Labs
- Universities
- Builders
- Investors
THOR open source coordination
- Data
- Models
- Hardware
Shared foundations, broader participation
Access and independence
Shared foundations give labs, manufacturers, researchers and operators more freedom to develop their own systems. Published interfaces and reproducible evaluation make it easier to adapt another team's work and test improvements independently.
Openness in practice
In practice, openness means publishing code, specifications and evaluation methods that others can inspect and improve. It also means releasing public datasets, model weights and hardware designs with clear rights to use and adapt them. More people should be able to contribute and build on that work.
These forms of openness can develop separately. Specifications and implementations each have their own release status. Contributor access follows participation rules, while private data remains governed by project permissions.
Releases and rights
THOR intends to develop these shared foundations in stages. Each release should make clear what is available and under which licence. Reproducing and adapting a model also depends on access to the relevant training code, data and usage rights. Private customer data and licensed technology remain subject to their own access and usage terms.
Participation explains permissionless annotation after the private beta. Data rights and access explains how public resources and private commercial work can coexist.