Video · Multimodal / AI & Machine Learning
Action-Conditioned Gameplay Dataset | 10,000 Hours of Video, Inputs & Game State for World Models
Advertised by providerLicensing rights not independently verifiedPrice on request
Overview
Action-Conditioned Gameplay Dataset | 10,000 Hours of Video, Inputs & Game State for World Models is a multimodal dataset. The provider states a volume of 10,000 hours. Typical applications include segmentation. Relevant sectors: AI & Machine Learning, Education and Media & Entertainment. The listing mentions timestamps, bounding boxes and polygons / segmentation. Pricing is not published and is provided on request.
Primary applications
What the listing offers
- Stated volume of 10k hours.
- Listing mentions timestamps, bounding boxes, polygons / segmentation.
Unlike passive gameplay videos, this dataset captures synchronized gameplay footage, player inputs, camera movement, game state, event logs, task goals, rewards, and outcomes, giving models the structure needed to learn how actions change an environment over time. Each trajectory includes rendered gameplay video, keyboard and mouse inputs, controller inputs, action timestamps, camera pose, view direction, player position, velocity, inventory state, health and status indicators, object interactions, environmental events, goal progress, reward signals, and success or failure labels. The dataset also includes depth maps, semantic segmentation, object bounding boxes, minimap state, NPC positions, collision events, physics state, and per-frame captions for supported environments. The dataset covers interactive game environments across first-person navigation, third-person action, platforming, driving, puzzle solving, resource collection, object interaction, inventory management, exploration, and multi-step goal completion. Trajectories are structured so models learn the relationship between the current visual state, the chosen action, and the resulting future state. Technical specifications include: • 10,000 hours of gameplay trajectories • Approximately 500,000 gameplay episodes • More than 1 billion rendered frames at source frame rate • More than 300 million timestamped keyboard, mouse, and controller actions • More than 50 million structured game-state snapshots, synchronized
Related datasets
More in Video & Physical AI →Video · Image · MultimodalAdvertised
Storyline Video Dataset - 82K Videos | Video Data
The Vertical Storyline Video Dataset is a large-scale collection of high-quality vertical videos featuring scripted and natural storyline-based content in various…
- Language
- —
- Volume
- 82k videos
- Domain
- Education +1
- Use case
- Evaluation & benchmarking
Price on request