Row transitions and headland turns
Isolate lane exits, turns, overlaps, and route recoveries across changing field geometry.
Independent infrastructure for agricultural robotics
Turn every field hour into data the next model can learn from.
Connect camera, LiDAR, GNSS, IMU, and operator events in one traceable loop—from field capture to model-ready dataset releases.
One operating record from field session to versioned dataset release.
MCAP + ROS
LiDAR + point clouds
Multi-camera video
Telemetry + trajectories
One operating loop
The field session is only the beginning.
Replace fragmented handoffs with a traceable path from synchronized field capture to the next model-ready dataset release.
- 01
Ingest synchronized field sessions
Bring MCAP, video, point clouds, GNSS, IMU, and machine events together without stripping away time or source context.
- 02
Curate the hard acres
Search and slice by field, route, crop stage, weather, intervention, and sensor condition to assemble the scenarios the next training cycle needs.
- 03
Build defensible ground truth
Combine model assistance with human 2D, 3D, and temporal review, then apply the quality gates each release requires.
- 04
Publish a versioned release
Freeze the manifest contents and source origins, attach declared rights metadata, then publish an immutable revision that the loader resolves for training and evaluation.
Scenario intelligence
Find the moments that change machine behavior.
Turn long field runs into focused training slices while preserving the sensor and operational context around every event.
Dust, glare, and low visibility
Compare perception behavior across dust, shadows, low light, occlusion, and changing crop density.
Operator interventions
Bring manual takeovers and recovery events into the next curation and review cycle.
Durable dataset identity
Ship a release your autonomy stack can reproduce.
Treat every training set as a governed artifact with a stable address—not a folder passed between teams.
- Friendly dataset reference plus an immutable canonical revision
- Manifest digest, object count, total size, and source origins
- Declared rights document and access terms attached to the revision
- Published state and revision identity returned through the loader
Dataset release manifest
Versioned- Dataset
- agriculture/row-navigation
- Release
- v18
- Origin records
- primary / avala · calibration / avala
- Declared rights
- Attached
- Revision state
- Published
agriculture/row-navigation@v18Built for field autonomy
One data system across the agricultural robotics program.
Keep navigation, crop-care, harvesting, and aerial-inspection work on the same dataset and quality foundation.
Autonomous field navigation
Curate row following, headland turns, obstacle avoidance, and route recovery from synchronized vehicle sessions.
Precision crop care
Build perception datasets for targeted spraying, weeding, plant detection, and crop-state understanding.
Harvest and manipulation
Review temporal interactions, object states, and difficult edge cases for picking and harvesting systems.
Aerial inspection
Turn drone imagery and telemetry into traceable datasets for field mapping, scouting, and inspection models.
Start with one model bottleneck
Map the field-data release your program needs next.
Bring your sensor stack, current workflow, and target behavior. Leave with a scoped path from field sessions to a versioned dataset.