Autonomy — self-driving fleet ops with a regulator-ready log

Autonomy is a white-label control plane for running a self-driving vehicle fleet — the operations and safety layer an AV program has to keep regardless of whose drive stack is under the hood. The read plane works before the backing actuation service is even deployed.

The vehicle registry:

Each vehicle is registered with its SAE automation level, its sensor suite, and the drive-by-wire stack vendor. This is the ground truth for what each car is actually rated to do.

The Operational Design Domain (ODD):

The ODD is a first-class safety envelope, not a note in a spec doc — geofence, speed cap, weather limits, road class, and a localization-quality floor. A vehicle is only allowed to drive itself inside its ODD.

Dispatching trips:

Trips are dispatched as passenger, delivery, repositioning, or validation runs. Every trip is bound to an ODD at dispatch time, so a car is never sent somewhere its envelope doesn't cover.

Downward-safe autonomy grants:

Autonomy is granted per shift and is downward-safe: a vehicle can drop itself to teleop on its own, but it can never self-promote to a higher autonomy level. The only direction a machine can move without a human is toward more caution.

Teleoperation handoff:

When a vehicle hands back to a remote operator, the platform tracks the live supervision ratio the program is actually staffed for — so you know at a glance whether you have enough operators for the cars currently leaning on them.

The disengagement log:

Every hand-back is recorded with its cause and a narrated reason — the disengagement log AV programs are expected to keep. Any real disengagement can be promoted into the scenario library as a replayable regression test against your current policy, so a bad moment on the road becomes a test you never fail the same way twice.