Worldmodeldata
About
About Worldmodeldata
Worldmodeldata captures information from inside a game engine, including each object's position, orientation and velocity. Controller and keyboard inputs are linked to the frames and state changes they produce, allowing researchers to train on relationships between actions and subsequent outcomes. The dataset also includes human annotations of goals, subgoals and planning.
Customers receive structured exports in Parquet, HDF5 or JSON-L for use in training pipelines. Framework agreements can cover additional game titles, custom property extraction, live engine instrumentation and specialised export formats. The company also offers controlled scenarios built from commercial games, enabling a research programme to specify the environments and behaviours it needs represented in the data.
Articles
Funding
- 2026-07-06 - Seed - $9.3M - Worldmodeldata raises GBP 7m seed round and emerges from stealth
Worldmodeldata raised GBP 7 million in seed funding led by Iona Star Capital as the Cambridge company emerged from stealth. - source: tech.eu
Investors
- Iona Star Capital - Lead - VC - 1 funding event - first seen 2026-07-06
Team
- Rhea Loucas - Founder and CEO
- Lord Richard Allan - Chairman
- Yann LeCun - Senior Science Advisor
Research
- Worldmodeldata publishes position paper on game-generated AI training data
2025-09-03 - Research Publication - The paper examines game-generated trajectories as a source of structured AI training data. It discusses Joint-Embedding Predictive Architectures and world-model learning. The work connects gameplay data with physical-AI, planning and self-supervised-learning research.
Source
Timeline
Events
- 2026-07-06 - Funding - Worldmodeldata raises GBP 7m seed round and emerges from stealth
Iona Star Capital led the GBP 7 million seed round. The financing supports product development, team expansion and additional licensed data-sourcing agreements. Worldmodeldata is building action-conditioned gameplay datasets for world models, physical AI and robotics companies.
Source - 2025-09-03 - Research Publication - Worldmodeldata publishes position paper on game-generated AI training data
The paper examines game-generated trajectories as a source of structured AI training data. It discusses Joint-Embedding Predictive Architectures and world-model learning. The work connects gameplay data with physical-AI, planning and self-supervised-learning research.
Source
Context
Comparable profiles
- Ropedia
stack adjacency - Physical AI Training Data
- LabyrinthAI
stack adjacency - Physical AI Training Data
- FORMOVE GmbH
stack adjacency - Physical AI Training Data
- Vision Lab
stack adjacency - Physical AI Training Data
- Sureform
stack adjacency - Physical AI Training Data
- Kognic
stack adjacency - Physical AI Training Data
- MiraiBit
stack adjacency - Physical AI Training Data
- JOYX
stack adjacency - Physical AI Training Data
- Realset AI
stack adjacency - Physical AI Training Data
- Innodata
stack adjacency - Physical AI Training Data
- H2L Robotics India
stack adjacency - Physical AI Training Data
- Xspark AI
stack adjacency - Physical AI Training Data
Shared workflows
- Physical Intelligence
workflow adjacency - Embodied World Model Training
- Reactor
workflow adjacency - Embodied World Model Training
- Stanhope AI
workflow adjacency - Embodied World Model Training
Context tags
- Embodied AI world models - Stack Layer - Official product and research material centres on world-model and embodied-AI training.
- Physical AI training data - Stack Layer - Worldmodeldata sells action-conditioned gameplay datasets specifically for physical-AI and robotics model training.
- Embodied world model training - Workflow - The synchronized gameplay stack is designed for training world models from action-conditioned trajectories.