Perception
Prepare camera and point-cloud ground truth for object detection, segmentation, and tracking models.
Independent infrastructure for automotive
Avala connects synchronized camera, LiDAR, radar, and IMU data with curation, human and AI review, quality control, and versioned releases. Autonomy teams can move from road event to training input with lineage intact.
One operating record from source session to dataset release.
urban-night / constructionIngest synchronized sessions
Curate the scenarios that matter
Build defensible ground truth
Publish a versioned release
One operating loop
Replace fragmented handoffs with a traceable path from multimodal capture to the next dataset release.
Bring MCAP, video, point clouds, and telemetry into synchronized sessions without stripping away source context.
Search and slice sessions for the edge cases, environments, and behaviors that belong in the next training cycle.
Combine model assistance with human 2D, 3D, and 4D review, then apply the quality gates each release requires.
Agree on the managed delivery format and source metadata. The self-service registry and public resolver remain in development; evaluate their availability separately from your current dataset handoff.
Durable dataset identity
Managed dataset operations are available today. This example shows the registry contract in development: an immutable manifest, revision, source origins and declared rights. Confirm the delivery format for your program before relying on resolver access.
autonomy/urban-night@v24Built for automotive AI
Keep perception, sensor-fusion, in-cabin, and mapping work on the same dataset identity and release discipline.
Prepare camera and point-cloud ground truth for object detection, segmentation, and tracking models.
Preserve timestamp and calibration context while teams inspect the same event across every sensor stream.
Organize visual and telemetry data for occupant- and driver-state models, with explicit review criteria.
Curate spatial data for mapping, localization, road understanding, and change-detection workflows.
Bring a real program
Bring one data flow, one release gate, and the stack it needs to connect to. We’ll show how Avala can run it without breaking lineage.