Sourcing field guide
Checking robot embodiment compatibility before sourcing
Compare hardware, sensors, and control assumptions before counting trajectories.
← Robotic Datasets
Original studio photography · Podcast Studio AZSource & media notes ↗
Fiund / Resources
The resource library.
Original guides to robotic datasets: source selection, collection design, annotations, evaluation, and project review.
27 guides shown
Sourcing field guide
What does a robot action vector actually mean?
Ask for semantics, units, frames, and execution context for every action field.
Sourcing field guide
Joint-space and end-effector action datasets
Clarify which robot representation your project can interpret and use.
Sourcing field guide
Coordinate frames in robot dataset briefs
Make spatial values interpretable across robot, tool, camera, and workspace references.
Sourcing field guide
Units and conventions that belong with robot data
Prevent silent interpretation errors in actions, state, and sensor records.
Sourcing field guide
Reviewing observation-action synchronization
Check which observation a robot command is paired with and how that pairing was created.
Sourcing field guide
Camera calibration and setup metadata for robot datasets
Request geometry information according to the model task and source configuration.
Sourcing field guide
Episode boundaries and resets in robot datasets
Keep attempts, setup, interruptions, and terminal conditions distinguishable.
Sourcing field guide
Defining success and failure in robot task data
Distinguish final outcome, procedural correctness, and observable evidence.
Sourcing field guide
Recovery demonstrations and retries in robot data
Preserve context around errors rather than retaining only final successful motion.
Sourcing field guide
What to ask about teleoperated robot demonstrations
Describe operator inputs, robot commands, control modes, and interventions.
Sourcing field guide
Understanding gripper fields in robot datasets
Clarify opening, closing, targets, feedback, and grasp evidence.
Sourcing field guide
Multi-camera observations in robot learning datasets
Define camera roles, stream identity, timing, and missing-view behavior.
Sourcing field guide
Grounding language instructions in robot episodes
Connect text to the requested task without confusing it with execution evidence.
Sourcing field guide
Object variation in manipulation datasets
Specify instances, properties, and task relationships that matter to the robot.
Sourcing field guide
Workspace and layout variation for robot datasets
Describe initial conditions and environment changes beyond a location label.
Sourcing field guide
Action horizons and sequence sampling in robot data
Keep recorded steps, training windows, and execution assumptions distinct.
Sourcing field guide
Sampling rates and missing records in robot datasets
Distinguish fresh measurements, resampled values, padding, and genuine inactivity.
Sourcing field guide
Reviewing sensor integrity in robot episodes
Check values, stream continuity, and task consistency before assessing model fit.
Sourcing field guide
Where human task video fits in robot learning data
Keep semantic demonstrations distinct from robot observations and action records.
Sourcing field guide
Combining robot datasets without losing source meaning
Standardize carefully while preserving embodiment, conventions, and provenance.
Sourcing field guide
Designing train and evaluation splits for robot datasets
Choose grouping rules that match the actual generalization question.
Sourcing field guide
Schemas and versions for robot dataset deliveries
Keep field definitions, episode structure, and revisions traceable.
Sourcing field guide
A practical pilot review for robot datasets
Resolve embodiment, timing, action meaning, and task evidence before expansion.
Sourcing field guide
Provenance questions for robot dataset sourcing
Keep collection history, transformations, and proposed use review attached to the data.
Sourcing field guide
Specify a robot demonstration dataset
Define embodiment, observations, actions, timing, and episode outcomes before sourcing.
Sourcing field guide
Human demonstration video versus robot action data
Understand which information each data type can contribute to an embodied AI project.