How to Evaluate a Self-Driving or Robotaxi Claim

How to Evaluate an Automated-Driving Claim

An automated-driving claim is incomplete until it names the system, version, vehicle, driving task, human role, fallback, operational domain, remote assistance, jurisdiction, permit, evidence, and current availability.

Start with what the person must do today. Do not start with the product name.

flowchart TD
    A["Read the claim"] --> B["Name system, version, vehicle, and date"]
    B --> C["Define driving task and human role"]
    C --> D["Define operational design domain"]
    D --> E["Define fallback and remote assistance"]
    E --> F["Verify permit and regulator state"]
    F --> G["Review exposure, interventions, crashes, and methods"]
    G --> H["State current availability without forecast inflation"]

Name the system before the level

Record the product name, software version, hardware, vehicle, model year, region, date, and whether the source describes a consumer feature, test fleet, pilot, employee service, invited service, or public service.

The same brand can operate supervised driver assistance, a development vehicle, a driverless test, and a ride service. Those are different systems and accountable roles.

Tesla's current FSD Supervised page says the feature requires an attentive driver and does not make the vehicle autonomous. That current instruction should control the consumer's role for that product.

Identify the dynamic driving task

The dynamic driving task includes lateral and longitudinal control, object and event detection, response, planning, and other functions needed to operate.

Ask which parts the system performs continuously. Ask which parts the person must perform, monitor, approve, or resume.

NHTSA's automated-vehicle guide says Level 2 provides steering and acceleration or braking assistance while the driver remains responsible and continuously monitors. NHTSA describes a mature automated driving system as performing the full dynamic driving task within a defined operational design domain.

The label should follow the task and human role. It should not override them.

State the human role in ordinary language

The person may be the driver, continuous monitor, fallback-ready user, passenger, safety driver, remote assistant, remote operator, fleet supervisor, or no participant in a particular task.

“Hands-free” does not mean eyes-free, driverless, or autonomous. “Driverless” in one service area does not mean universal operation.

The current manual and terms matter more than the marketing shorthand. A consumer should follow the vehicle's official instructions and law. This guide does not authorize use or reduced attention.

Define the operational design domain

The operational design domain identifies where and when an automated driving system is designed to operate.

NHTSA's Vision for Safety identifies roadway type, geography, speed, weather, light, and other conditions as domain elements.

A complete domain can also include maps, construction, traffic, emergency scenes, connectivity, road markings, pickup zones, vehicle state, and fleet-support limits.

“Works in Austin” is not a domain. Name the service area, roads, times, weather, speed, rider access, vehicle, supervision, and exclusions.

Explain fallback

Fallback answers what happens when the system reaches a limit, detects a fault, leaves its domain, or cannot complete the driving task.

In driver assistance, the attentive driver may remain responsible throughout. In a Level 3 design, a fallback-ready user may receive a takeover request. In a driverless Level 4 service, the system and operator need a minimal-risk response within the domain.

Remote assistance must be described. A remote person can provide route guidance or exception help without continuously driving. Direct remote control is a different mode.

The claim should name who is responsible at the moment of failure.

Verify the regulator state

A permit is not one universal approval.

California DMV's permit resources separate safety-driver testing, driverless testing, and deployment. Other jurisdictions use different authorities.

Record the agency, permit category, vehicle, operator, domain, effective date, conditions, reporting, and whether passenger service requires another authority.

A private-site demonstration, federal disclosure, company filing, or consumer driver-assistance release is not automatically a local driverless deployment permit.

Review the safety evidence

Safety evidence needs exposure and method. Record miles or hours, domain, traffic mix, weather, vehicle count, software version, interventions, disengagement definitions, crashes, severity, fault method, reporting threshold, and comparison group.

NHTSA's Standing General Order page explains current crash reporting for Level 2 ADAS and Level 3 through 5 ADS. It also warns about data quality, classification, duplicate reports, and comparability.

A crash count without exposure is not a rate. A rate without domain and method is not a fair comparison. Company-selected metrics should remain attributed.

NHTSA's Voluntary Safety Self-Assessment index can reveal company disclosures. Inclusion is not federal endorsement or approval.

Separate present access from forecasts

Use a state key: concept, prototype, closed-course test, safety-driver test, driverless test, bounded pilot, limited public service, expanded service, consumer feature, or forecast.

Tesla's 2025 Form 10-K says the company launched a Robotaxi service with Model Y vehicles in June 2025 and expects Cybercab later. That company filing should not be rewritten as universal Cybercab availability.

Verify the current service terms, geography, vehicle, human role, remote assistance, regulator, customer access, and date.

Complete the claim

A complete statement might say that a named software version provides Level 2 driver assistance on a specified vehicle and requires a continuously attentive driver under the current manual.

Another might say that a named operator runs a driverless Level 4 passenger service within a specified domain under a named permit, with remote assistance and a documented minimal-risk response.

If those fields are missing, the claim is not ready for a safety conclusion.

This guide was developed with AI assistance from the immutable E039 transcript and linked NHTSA, California DMV, Tesla, regulator, and automated-driving framework records. Dalton Anderson remains the author. Continuous vehicle-safety, legal, regulatory, current-source, and founder review are mandatory before publication. Publication is not authorized.