Understand everyday actions
Specify the tools, objects and environments your model needs to observe. First-person sequences can retain the context around a decision or movement.
FIUND / Physical & embodied AI
Data grounded in actions, objects and environments.
Scope first-person video, practical tasks and physical-world data for models that need to understand human activity. Start with the perspective and sequence your task requires, then evaluate a suitable source or capture plan.
FIUND / COLLECTION STRUCTURE
StreamReference
Depth / spatial representationShared timebase
Motion / positionShared timebase
Image / event contextShared timebase
ILLUSTRATIVE FORMAT Assets confirmed per collection
Model tasks
Choose data for the behavior you need.
Interpret skilled work
Define the procedure and the steps between problem and outcome. Narration, camera placement and continuous coverage serve different learning goals.
Relate physical-world streams
For sensor-based work, define the time reference, coordinate system and calibration information required. These are project requirements to verify, not automatic properties of every source.
Relevant data types
Explore the source formats and context that could support your task. Access and suitability are confirmed for each proposed project.
Scope before scale
Define a sample you can evaluate.
Capture perspective
Camera position, field of view, task coverage, participant requirements and environment.
Sequence and outcomes
Task boundaries, interruptions, successful and unsuccessful attempts, and surrounding context.
Sensor references
Required modalities, calibration, synchronization and coordinate-system documentation.
Sample review
Questions to settle early
- The task is visible from the selected perspective
- Sequences preserve necessary context
- Required sensor references are demonstrated
- Capture and participant permissions fit the use
Start with a scoped brief.
Tell us the source, coverage and intended use. We assess fit and permissions, identify preparation needs, and agree a pilot or delivery scope.
Frequently asked questions
Is every listed physical-world category available immediately?
No. These pages describe the data we work with. Source fit, access, permissions and any new capture are confirmed for each brief.
Can FIUND scope a custom capture?
Share the task, environment, perspective and participant requirements. A proposed pilot can establish feasibility and acceptance criteria before a larger collection is agreed.
Do egocentric recordings include motion or depth sensors?
Not by default. List the sensor streams and reference information your project needs so the proposed source or capture setup can be assessed.