For Physical AI teams

Tell us what your next training dataset needs.

Discuss labeling existing recordings, collecting new demonstrations, or evaluating dataset access. Start with your sensors, tasks and required outputs. Together, we can identify the relevant workflow, open questions and acceptance criteria before agreeing on an evaluation or a production program.

Book a working session Explore the Platform

How should you evaluate a data program?

Use this checklist to scope a benchmark with Avala or another provider. Agree on evidence and acceptance criteria before work starts. Checking a box records your review in this page only; it does not certify a provider or submit a request.

Define the output contract: sensors, coordinate frames, timestamps, labels, actions, measured state and episode boundaries.
Test the training path: load a representative export into your actual training loader and check for missing fields.
Measure useful yield: report accepted episodes or hours after technical review and deduplication, broken down by coverage.
Audit quality: compare against an independent reference sample, including occlusions, track identity and action boundaries.
Compare accepted cost: include integration, calibration, review and rework in the cost per accepted deliverable.
Evaluate the model: compare against your baseline on held-out conditions using the same evaluation protocol.
Close the loop: trace a failure through review to a new release. Confirm ownership, portability and exit terms in writing.

Blank planning templates in English. Define requirements and record evidence before agreeing on a program. These files contain no results or commercial terms.

Your working session

Book a working session

Your use case, sensor types, output schema and acceptance criteria. Keep recordings, bucket credentials and confidential project details private until an appropriate sharing process is agreed.

What do you need?

Pick one and the session is scoped before it starts.

What we will cover

A working session built around your data loop.

We will use your real modalities, constraints, and model objective, not a generic capabilities tour.

Map the data-to-model loop

Trace ingest, curation, annotation, QC, export, training feedback, and re-label around your real workflow.

See the system of record

Walk through Mission Control with your modalities, quality thresholds, and deployment constraints in view.

Define the first release

Discuss a representative input, the required output and how to evaluate it. Agree on scope, responsibilities and commercial terms separately before work begins.

Mission Control

One operational record from sensor to model.

See how raw scenes, annotation decisions, quality evidence, dataset versions, and model feedback stay connected instead of disappearing across hand-offs.

Start with the real bottleneck

Leave with a scoped path forward.

Bring your data requirements. Use the conversation to identify fit, unresolved questions and a practical next step. A scoping conversation does not establish production quality or model improvement.

Book a working session

Prefer email? Reach us at [email protected] and we'll coordinate a session.

Existing customer? Sign in to Mission Control for support and success resources.