DatoricAction-Conditioned Gameplay Dataset
Available · action-conditioned trajectories
Train world models and agents on gameplay trajectories that align rendered video with player inputs, camera motion, state changes, events, rewards, goals, and outcomes.
Dataset specification
- Collection capacity: 1B+ rendered frames
- Collection capacity: 300M+ timestamped input events
- Collection capacity: 50M+ state snapshots
- Video
- Input logs
- Camera motion
- Player and object state
- Event labels
- Reward labels
- Success/failure outcomes
- Sync checks
- JSON trajectories
Intended uses
- Action-conditioned world-model training
- Agent policy and inverse-dynamics learning
- Interactive video and future-state prediction
- Goal, reward, and outcome modeling
Scope notes
- Environment, title, input-method, state-field, episode-length, and frame-rate distributions are confirmed for the selected delivery.
- Depth, segmentation, object, minimap, NPC, collision, physics, and caption fields are subset-specific and identified during review.
- Repository metadata illustrates the trajectory schema; real gameplay trajectories are supplied in the buyer review package.
Collection scope
Environment coverage
- First-person navigation and third-person action
- Platforming, driving, puzzle solving, and resource collection
- Object interaction, inventory, exploration, and multi-step goals
Trajectory scale
- Collection capacity of 1B+ rendered frames
- Collection capacity of 300M+ input events and 50M+ state snapshots
Available scale
- Available program
- Synchronized video, input, state, event, and outcome streams
Buyer review package
- Trajectory schema, data dictionary, and sample metadata in the repository
- Real trajectories with video, input logs, state logs, and QA summaries
- Target environments, input methods, state fields, and hours confirmed for the selected delivery
Annotation & metadata fields
- Action timestamps and keyboard, mouse, or controller inputs
- Camera pose and view direction
- Player position, velocity, inventory, and health or status
- Object interactions and environmental events
- Goal progress and reward signals
- Success and failure outcomes
- Episode, frame-rate, input-frequency, and state-frequency metadata
- Optional depth, segmentation, object, minimap, NPC, collision, physics, and caption fields
Capture methodology
- Rendered gameplay video synchronized with keyboard, mouse, or controller inputs
- Camera motion, player state, object state, environmental events, goals, rewards, and outcomes aligned by trajectory
- Episodes structured as state, action, and resulting future-state sequences
- Subset-specific depth, segmentation, object, physics, and caption signals recorded where available
Provenance & rights chain
- Rights-clearance and chain-of-custody documentation included in licensing review
- Commercial license issued directly by Datoric
- Environment, capture, input, state, and episode lineage recorded for the selected delivery
- Collection and annotation activity handled under Datoric's published privacy notice
Quality, duplicates & PII
How submissions are reviewed and cleaned before they are accepted into the dataset.
- Action-state synchronization, frame-rate, input-frequency, and state-frequency checks
- Goal-completion, reward-signal, and metadata-consistency review
- Corrupted and malformed episode filtering
- Duplicate-episode removal and rights-clearance review
Formats & delivery
- Gameplay video
- Input, state, and event logs
- CSV metadata
- JSON trajectory files
Rights & license scope
Licensed directly by Datoric for commercial AI training, with final scope controlled by the signed agreement for the selected delivery.
Version & verification
- Availability
- Available
- Datasheet version
- July 21, 2026
- Release date
- July 21, 2026
- Last verified
- July 21, 2026
- Owner
- Datoric
Frequently asked
- How is this different from passive gameplay video?
- Each trajectory aligns rendered video with player inputs, camera motion, game state, events, goals, rewards, and outcomes so models can learn state-action transitions.
- Which state fields are included?
- Core fields can include camera pose, player position and velocity, inventory, health or status, object interactions, events, goal progress, rewards, and outcomes. Exact coverage is confirmed for the selected subset.
- Can we inspect complete trajectories?
- Yes. The buyer review package includes real video, input logs, state logs, QA summaries, and licensing documentation for selected environments.
- Can Datoric target a different environment or event schema?
- Yes. Custom collection can target environments, input methods, state and event fields, goals, reward structure, and evaluation criteria.