A08Labs

Self-models for machines and robots

Machines change. Their models should too.

A08 detects changes in machine dynamics and supports verified local repair during operation.

Action-conditioned predictionLocal repairIndependent verification

Same-run repair demonstration

A fault appears. The mission continues.

An impact reverses the commands sent to the robot's left leg. The frozen response falls. A08 identifies the changed mechanism, verifies a local correction and applies it without resetting the run.

0.22 sdetect
0.03 scompile
4 / 4recovered
0resets
Inspect the result
Independent MuJoCo simulation using the public Unitree G1 model and policy from Unitree RL Gym. No hardware result or Unitree endorsement is implied.

Production line · Simulation

One robot slows. The line adapts.

Two arms, one conveyor. When the packing robot loses 30% of its motion response, A08 repairs the line model and adjusts the conveyor and first robot's timing.

Same five boxes. Same slowdown. Original model first, A08 repair second. Independent ManiSkill / SAPIEN simulation using xArm6 Robotiq assets. No physical validation or manufacturer endorsement is implied.
3 / 5Original schedule
5 / 5A08 repair + replan
Explore the use case

The operating problem

Most machine models become stale the moment the hardware changes.

01

Wear changes response

Actuators, joints and contact surfaces stop behaving like their nominal model.

02

Context changes dynamics

Payload, temperature, friction and latency alter the effect of the same command.

03

Broad retraining is slow

It changes more than the affected mechanism and is difficult to verify during operation.

The A08 system

A repair layer between observation and action.

A08 does not replace the controller. It keeps the predictive model behind the controller aligned with the machine.

Model

Machine self-model

Predicts future state from recent observations and candidate actions.

Monitor

Competence-gap detector

Flags consequential differences without reading hidden mechanism state.

Repair

Local patch compiler

Turns bounded evidence into a small, attributable model correction.

Govern

Independent verifier

Checks improvement, regressions, limits and provenance before routing.

One product direction

Start with repair. Build the network from deployed machines.

Each deployed self-model is tied to an evidence record containing its exact version, repair lineage, evaluations and operating limits.

Explore the technology
01Repair Runtime

Detect, diagnose, verify and route bounded changes around one machine.

02Evidence Registry

Keep every model, repair, evaluation and rollback path attributable.

03Machine Network

Distribute proven self-models and repair modules across compatible fleets.

Design partners

Built for teams operating real machines.

Robot OEMs

Make adaptation part of the platform

Add verified model repair across changing payloads, actuators and operating conditions.

Autonomy teams

Know when the planner is wrong

Connect prediction error to a local mechanism and a controlled recovery path.

Fleet operators

Reduce diagnosis and downtime

Turn repeated field changes into reusable, versioned machine knowledge.

Pilot A08

Bring one machine and one change that matters.

We are looking for robotics partners with hardware access, telemetry and a repeatable mission.

Clear consent, no purchased lists. Read our privacy notice.