AuraOne

Product

Robotics data, built to be audited.

Collect trajectories with session durability, quality scoring, and exportable artifacts. Start in simulation, graduate to hardware when you are ready.

Teleoperation

Session lifecycle + telemetry capture in Postgres (durable state for sessions, collisions, e-stops, and trajectories).

Human demonstration

Ingest and normalize human demos into trajectory records that can be quality-scored and exported.

VLA data

Pair vision, language instructions, and action sequences with timestamps and metadata.

Safety-first posture

Quality scoring and telemetry help teams detect unsafe trajectories early. Hardware safety controls remain external and program-specific.

Robotics LabTeleopSim AdaptersExports
Trajectory scoring

Safety and smoothness scores are computed from telemetry signals (collisions, e-stops, duration) with reasons attached.

safetyScore: collisions + e-stops penalties
smoothnessScore: short-duration + collision proxy
overallQuality: weighted blend + success signal
Simulator boundaries

The repo implements adapter interfaces for Isaac Sim, PyBullet, and MuJoCo so pipelines can be exercised and typechecked without bundling vendor runtimes.

“Integration boundary” means the interface exists. Real simulator runtime deployment is external.
How it works
  1. Capture: Create sessions and store trajectory pointers (video/sensor/command URLs).
  2. Score: Compute safety and smoothness proxy metrics with explainable reasons.
  3. Simulate: Exercise adapters for Isaac Sim, PyBullet, and MuJoCo.
  4. Export: Export datasets as JSON payloads tagged as RLDS/OpenX for partner handoff.
Truth check

Data models, helpers, scoring, and export logic exist in the repo. Real-time media streaming, VR controllers, and partner hardware pilots are external rollouts.