From Robotaxi Pilots to Regulated Deployment: U.S. Autonomous Vehicle Rules in 2025–2026

Overview

The central U.S. autonomous-mobility question is no longer only whether a vehicle can drive without a person at the controls. It is also whether that vehicle can enter public service, remain observable, respond to authorities, and be corrected or removed when its operation becomes unacceptable. A demonstration establishes a bounded capability; commercial deployment adds continuing legal and operational obligations.

Update note: This September 2026 revision replaces the 2025 short-term robotaxi outlook with a source-based analysis of the regulatory conditions for sustained driverless deployment. Earlier measures are retained only where they explain the framework in force today.

California authorized fare-charging autonomous passenger-service programs in 2020, and NHTSA updated occupant-protection standards for ADS-equipped vehicles without conventional manual controls in a final rule published on March 30, 2022 and effective September 26, 2022. The 2025-2026 change is therefore not a move from no regulation to regulation; it is greater specification of how continuing deployment is authorized, observed, and corrected. ^{[1] [24]}

In this article, regulatory maturity refers to observable functions such as identifiable operators, defined permissions, reportable events, emergency interfaces, and corrective processes. It does not imply that one regulatory model is superior or that autonomous vehicles have already achieved a particular safety outcome.

The analysis focuses on federal vehicle-safety oversight, California and Texas implementation, and Arizona as an earlier state-centered comparison. It is not a fifty-state legal survey, and it does not analyze Level 2 driver-assistance products as if they were driverless services. ^[10]

The structural conclusion is that driverless deployment requires a maintained chain of permission and accountability, not a single approval event. Technical comparisons among Cybercab, Waymo, and Uber are outside the scope of this regulatory analysis.

What Changed: From General Direction to Operational Obligations

The starting point was already a layered system

Several important federal measures were already in place by September 2025. DOT announced an AV framework in April 2025. ^[2] NHTSA's third amended crash-reporting order became effective that June. ^[11] Zoox received a domestic demonstration exemption in August. ^[3] These measures are the baseline for assessing what became more operational during 2026.

During 2026, California adopted revised testing and deployment regulations, and Texas implemented commercial driverless authorization. ^{[16] [26]} At the federal level, NHTSA granted a commercial exemption for a purpose-built passenger vehicle, withdrew the proposed AV STEP program, and announced additional standards work. ^{[6] [7] [9]} September then added a national strategy and a compliance inquiry into Tesla's Cybercab certification. ^{[12] [13]}

Maturity is not the same as a longer rulebook

A useful distinction is therefore between legal specification, administrative implementation, and demonstrated outcomes. Legal specification identifies an obligation. Implementation makes it possible to submit the required material, contact the responsible organization, or enforce the obligation. Outcomes require evidence about what happened after those arrangements operated. These stages are analytically distinct and cannot be collapsed into one announcement-driven claim.

California's revised reporting mechanism was in force at the review cutoff, but its first quarterly cycle had not yet produced a completed dataset. ^[21]

The remainder of the article organizes the evidence by function rather than by press-release date. The causal sequence is: establish the vehicle's legal basis, authorize the relevant activity, identify responsible parties, observe operation, handle abnormal situations, and apply correction when necessary. That sequence also makes clear where the system remains incomplete.

The U.S. Autonomous Deployment Stack

There is no single U.S. regulator whose permission automatically settles every question about a robotaxi. The federal strategy distinguishes federal vehicle and equipment safety responsibilities from state and local road-use responsibilities. California separately administers autonomous-vehicle permissions and passenger-service authority through different agencies. Arizona, meanwhile, expressly limits additional local AV-specific regulation while preserving specified airport powers. ^{[14] [24] [31]}

The following Autonomous Deployment Stack is this article's analytical model, not the name of an official government program. It separates decisions that are frequently blended together in public discussion. The functions can overlap, and the institutional allocation differs by jurisdiction; they are not five compulsory applications filed in an identical order in every state.

Different regulatory functions move on different timelines

Vehicle standards, state operating authority, continuing oversight, and passenger-service authority can change on separate schedules even when they concern the same fleet. A proposed FMVSS amendment, an effective state authorization, a reporting deadline, and a commercial-service permission therefore describe different legal functions. For longitudinal analysis, the useful question is which regulatory function changed at a given date, rather than whether autonomous-vehicle policy as a whole became more or less mature. ^{[6] [21] [26]}

Permissions are complementary, not substitutes

Federal vehicle compliance, state road-operation authority, and passenger-service permission answer different questions. Testing authority also does not automatically confer commercial deployment rights. When assessing a launch, the useful object is the specific permission: the covered vehicle or system, responsible legal entity, authorized use, operating boundaries, conditions, and duration.

The operational design domain is a boundary, not a promise

California's adopted definitions describe the operational design domain, or ODD, through the conditions in which an ADS is designed to operate. These can include environmental conditions, geography, roadway characteristics, and time-related limitations. The regulatory significance is the boundary around the claimed capability, rather than a suggestion that the vehicle can perform the same task everywhere. ^[18]

As an analytical consequence, expansion into a new environment raises two separate questions. Is the system capable of operating there, and does the relevant permission cover that operation? The answers can diverge. An engineering team may believe the system is ready before the necessary authorization changes; a broad legal pathway may exist before a particular system can use it responsibly.

Autonomous mobility therefore couples technical capability with an institutional boundary: regulation does not supply perception or planning, but it defines where that capability may be used and which organization remains accountable for the operation.

Vehicle Standards, Self-Certification, and the Exemption Bridge

Separating the safety objective from the human interface

A purpose-built driverless vehicle can create a regulatory problem before any debate about the quality of its driving model. Requirements written around a human driving position, a pedal, or a display may need interpretation or modification when that interface is absent. The challenge is not simply whether a rule is old. It is whether its safety objective can be maintained without the physical assumption around which the requirement was written.

The federal government had already addressed part of this problem in its 2022 occupant-protection final rule. That change concerned crashworthiness requirements for ADS-equipped vehicles without traditional manual controls. It did not eliminate the entire body of Federal Motor Vehicle Safety Standards, or FMVSS, and it did not create a general approval of autonomous driving competence. ^[1]

NHTSA's 2026 modernization effort is broader than two standards. It proposed updates covering transmission shifting (FMVSS 102), windshield defrosting and wiping (103/104), tire placards (110), and braking systems (135). It also announced plans to update controls and displays (101), lighting (108), mirrors and rearview display (111), electronic stability control (126), and sun visors and warning labels (201/208). The March 102 and June 135 actions were proposals; the July 30 announcement distinguishes completed proposals from additional planned updates. ^{[4] [5] [7]}

The reusable reasoning is a form of regulatory translation: identify the original safety function, identify the human-interface assumption, and determine how the function is to be demonstrated in the new configuration. Removing a pedal requirement and removing a braking-performance requirement are not the same change. Equally, removing a display intended for a human operator does not establish that the automated system makes correct driving decisions.

Self-certification is not regulatory preapproval

The U.S. vehicle-safety system relies on manufacturers certifying compliance with applicable standards, subject to federal oversight. NHTSA's explanation of that model distinguishes manufacturer certification from an agency-issued preapproval of each vehicle. Its longstanding interpretation also emphasizes reasonable care and the manufacturer's responsibility for compliance when the agency tests the product. ^[32]

This arrangement assigns initial compliance responsibility to the manufacturer while preserving federal review. NHTSA opened Audit Query AQ26002 on September 3, 2026, the day Tesla began commercial Cybercab service in Austin, and publicly announced the inquiry on September 4. ^[12] On September 10, NHTSA's Chief Counsel issued a Special Order requiring Tesla to provide sworn responses to 21 requests by September 30. The order states that failure to respond fully or truthfully may subject Tesla to civil penalties of up to $27,874 per day, capped at a statutory maximum of $139,356,994 for a related series of daily violations. That figure is the maximum stated in the order, not a penalty NHTSA has imposed or announced that it is seeking. ^[34]

The Special Order seeks the technical and procedural basis for Tesla's certification, including which FMVSS Tesla considers applicable, the role of any temporarily attached human controls, and compliance questions involving FMVSS 101, 102, 108, 111, 126, and 135. It remains an investigative demand rather than a final finding that Cybercab is noncompliant. ^[34]

Demonstration and commercial exemptions are different instruments

The Automated Vehicle Exemption Program, or AVEP, and the general temporary-exemption process under Part 555 are not interchangeable permissions. NHTSA's August 2025 Zoox action was a domestic demonstration exemption following the April expansion of AVEP. The agency also required removal or covering of labels asserting complete FMVSS compliance. That episode concerned a particular exemption pathway, not an unrestricted authorization to commercialize any driverless design. ^{[2] [3]}

The July 2026 commercial grant was a different action. NHTSA exempted the Zoox sedan from portions of eight standards: FMVSS 103 (windshield defrosting/defogging), 104 (wiping/washing), 108 (lamps and associated equipment), 111 (rear visibility), 135 (light-vehicle braking), 201 (interior-impact protection), 205 (glazing), and 208 (occupant crash protection). The exemption runs through July 31, 2028 and permits no more than 2,500 exempt vehicles to enter interstate commerce for commercial deployment in any twelve-month period. ^[6]

Operational Authorizations are separate NHTSA letters issued under the exemption's oversight condition. They prescribe where and under what operating conditions exempt vehicles may run, can limit the number operating simultaneously, and may be modified or expanded as the operation changes. The 2,500-vehicle annual exemption ceiling therefore does not by itself authorize all 2,500 vehicles to operate at once or in every location. ^[6]

NHTSA had outlined a more adaptable oversight approach in its June 2025 Part 555 letter. The 2026 grant applies that approach through operating conditions that can be revised as the vehicle's use changes, while maintaining specified grant conditions. An exemption therefore can be both an entry mechanism and a continuing oversight relationship. ^{[6] [33]}

The exemption as a transitional bridge

A temporary exemption can address a specific mismatch while permanent standards evolve, but it remains bounded by the grant and its operating conditions. Scaling beyond those bounds can require a revised authorization, a different compliant design, another exemption step, or final rulemaking. The value of the mechanism is therefore precision about what may operate now and what legal change would be needed for a different operation.

Federal Oversight: A Withdrawn Program and a Continuing Feedback Loop

AV STEP did not become the national supervisory layer

NHTSA announced the proposed ADS-equipped Vehicle Safety, Transparency, and Evaluation Program, known as AV STEP, in December 2024; the proposal followed in January 2025. It contemplated a voluntary federal framework for evaluating participating ADS operations and increasing transparency. That proposal is relevant as an alternative institutional design, not as a program that companies were already legally required to join. ^[8]

On June 26, 2026, NHTSA withdrew the proposal. The withdrawal discussed reassessment of the proposal, public comments, and other ADS initiatives. It also recorded different objections: some industry commenters questioned the burdens and incentives of participation, while safety and local-government organizations questioned whether a voluntary approach supplied sufficient oversight or local involvement. Those positions explain a contested policy choice; they do not establish that either set of concerns was conclusively resolved. ^[9]

The narrower, supportable conclusion is that this proposed additional program did not proceed. The withdrawal did not itself repeal the agency's existing standards, exemption, reporting, or defect authorities. The subsequent commercial exemption and ongoing certification inquiry illustrate other instruments continuing to operate. It would therefore be inaccurate to describe the change either as the completion of a new national approval system or as the disappearance of federal AV oversight. ^{[6] [12]}

Reporting changed in both scope and burden

The federal Standing General Order originated in 2021. Its third amendment, issued in April 2025 and effective June 16, revised the obligations of specifically identified manufacturers and operators. It differentiated ADS and Level 2 reporting, generally required reports of specified serious crashes within five days after the reporting entity receives notice, and used monthly reporting for certain less severe events. It also reduced duplicative reporting, updates without new information, and monthly submissions with nothing reportable. ^[11]

The amendment both retained safety-critical reporting and streamlined some duplicative or empty submissions, so the change is better described as a revision of scope and burden than a simple increase in reporting.

The order is important because it provides a route from an observed event to a regulatory question. NHTSA describes using reported information to identify matters requiring follow-up and, where warranted, investigations or recalls. A report is not itself a determination that the ADS caused the crash. Initial information can be incomplete, and an investigation may need additional evidence. ^[10]

The resulting analytical loop is: operation produces evidence; evidence can trigger inquiry; inquiry can establish a need for correction; correction changes subsequent operation. Not every report moves through every stage. An institution capable of receiving data but unable to act on a demonstrated problem would have an incomplete loop. An institution acting without adequate evidence would face a different weakness. Reporting and enforcement therefore perform complementary functions.

Incident counts cannot carry conclusions they were not designed to support

NHTSA cautions that its public crash counts are not normalized by miles traveled or operational design domain. Knowledge of incidents can differ with telemetry and reporting access, and the same crash can appear in more than one report. These limitations prevent a direct inference that the entity with fewer reported crashes has the safer system. ^[10]

The methodological consequence extends beyond company comparisons. A change in reported incidents after a new rule could reflect changed reporting thresholds, increased exposure, better information collection, different conditions, or actual behavioral change. A defensible evaluation needs a way to separate those possibilities. The establishment of a reporting obligation is thus evidence of observability, not a shortcut to a causal safety conclusion.

California: Making Driverless Operation Administratively Specific

A revised framework, not the beginning of permission

California's April 28, 2026 adoption of revised AV regulations is one of the clearest state-level changes in the review period. The package revised testing and deployment oversight and created a pathway for autonomous trucks and transit vehicles. Individual operations still require the permissions applicable to their activity. ^{[16] [17]}

For passenger vehicles, the important structure remains the distinction between testing with a driver, driverless testing, and deployment. The adopted texts attach requirements to those different activities rather than treating autonomy as a single undifferentiated permission. They also make the operating domain and supporting safety case part of the regulatory record. ^{[18] [19]}

A safety case links an operating claim to supporting arguments and evidence that the authority can examine and revisit. Its presence makes the operating basis more explicit without, by itself, proving future safety outcomes.

The missing driver leaves several different responsibilities

A human driver normally combines vehicle control with communication, response to police instructions, and assistance during unexpected situations. In a driverless service those functions cannot simply be assigned to the word autonomy. The relevant organization needs to know which responsibilities remain with the ADS, which require remote personnel, and which require an on-site response. That is a problem of responsibility allocation before it is a comparison of technical implementations.

California's adopted definitions expressly distinguish remote assistance from remote driving. Assistance provides information, advice, or revised goals without performing the real-time driving task. Remote driving involves real-time performance of that task, such as steering or braking. Those are different activities even when both involve someone away from the vehicle. ^[18]

Driverless mobility therefore still requires an accountable operating organization around the vehicle. Human responsibilities may be redistributed among software, remote personnel, fleet operators, and on-site responders, but they do not disappear merely because there is no licensed driver in the front seat.

Emergency interaction becomes a defined interface

AB 1777 was enacted in 2024, but key requirements entered operation in 2026. Its chronology therefore illustrates why enactment date and implementation date are analytically distinct. For covered vehicles, the law establishes a communication interface with emergency personnel and supports processes for directing autonomous operations around emergency areas. These are requirements for interaction with public authorities, not a general judgment about driving performance. ^[23]

The adopted deployment text specifies, for the relevant covered operations, response to an emergency contact call within thirty seconds while vehicles are on public roads. It also requires fleet directions responding to an emergency geofencing message within two minutes. That second requirement concerns issuing directions; it does not mean that every vehicle must physically evacuate the area within two minutes. ^[19]

These deadlines make emergency coordination testable at the organizational level: authorities have a defined contact path and the fleet has a defined response obligation. The two-minute rule concerns issuing fleet directions, not guaranteeing that every vehicle completes a physical maneuver within two minutes.

Traffic noncompliance is given a reporting destination

Beginning July 1, 2026, California's implemented noncompliance process allows law enforcement to issue notices concerning alleged traffic violations by covered autonomous vehicles. The DMV's memorandum directs manufacturers to transmit notice information within seventy-two hours of receipt, or twenty-four hours for priority notices. Failure to meet reporting obligations can lead to administrative action. ^[22]

The notice process gives an alleged violation an organizational recipient and a route into the DMV record. It addresses regulatory reporting and administrative accountability; allocation of civil liability is a separate legal question.

Reporting extends beyond the collision headline

The DMV now supplies permit-specific templates for collisions, vehicle immobilizations, driving-task-relevant system failures, vehicle miles traveled, and, for applicable testing reports, braking events. The reporting memorandum sets different schedules and requirements by permit category. It would be inaccurate to say every operator files every template on the same timetable. ^{[20] [21]}

The implementation schedule also matters. Monthly reporting began on August 26, 2026, while the first quarterly cycle had not yet closed as of September 21. Accordingly, the article treats the reporting design and submission machinery as observed developments, not as a completed dataset proving the regulations' effectiveness. ^[21]

California's reporting design broadens observability beyond major crashes to events such as immobilizations and system failures. Interpretation of those records will still depend on exposure, definitions, and the quality of the underlying data.

A Road-Legal Vehicle Is Not the Entire Passenger Service

California's two permissions answer different questions

California's division between the DMV and the California Public Utilities Commission, or CPUC, makes the service layer unusually visible. The CPUC established AV passenger-service pilot programs in 2018 and fare-charging deployment programs in 2020. Participation requires the corresponding DMV authority, and driverless programs include passenger-safety-plan requirements. Commercial passenger authority is therefore neither identical to vehicle compliance nor a new invention of 2026. ^[24]

The distinction is useful even outside California as an analytical question: what is being authorized, a vehicle's operation or the transportation service offered to passengers? A vehicle can satisfy a narrowly defined road-operation requirement without answering every service question. Conversely, an established transportation platform does not settle the legal status of the automated vehicle that supplies a ride.

The CPUC's August 28, 2025 rulemaking announcement expanded the questions under consideration to include new partnerships and business models, personally owned autonomous vehicles, airport operations, shared rides, and unaccompanied minors. That announcement establishes the scope of a proceeding. It does not establish that each listed model has been approved or that a final rule has resolved every issue. ^[25]

The institutional unit can extend beyond the vehicle

The significance of these questions is structural. Once several organizations participate in delivering a trip, it becomes important to distinguish who accepts the passenger, who supplies the vehicle, who operates the ADS, who receives a complaint, and who must act in an emergency. A contractual relationship may divide tasks, but an analyst cannot infer the regulatory allocation merely from the brand displayed in an application.

Passenger access and passenger safety are also not reducible to collision avoidance alone. Consider, as analytical questions rather than statements of a particular rule, how a person requests help, how a trip is ended during a disruption, and which organization handles a service failure. These questions explain why autonomous transportation regulation extends beyond permission for an ADS to control steering and braking.

Technical comparisons among Cybercab, Waymo, and Uber are outside this article's scope. Their vehicle intelligence, fleet operations, and service-distribution models can differ even when they participate in the same broader autonomous-mobility market.

Texas and Arizona: State-Centered Rules With Different Entry Mechanisms

Texas added an enforceable commercial authorization

Texas provides a second implementation case rather than a simple opposite to California. Senate Bill 2807 became effective on September 1, 2025. The latest implementing rules became effective on February 27, 2026, and the commercial authorization requirement became enforceable ninety days later, on May 28. TxDMV opened its application process in April and subsequently announced implementation of enforcement and complaint processes. ^{[26] [27] [28]}

The scope matters: the program covers driverless transportation of passengers or property in furtherance of a commercial enterprise. Individually owned vehicles used personally do not require this authorization. The fact that a vehicle has an advanced assistance system does not, without examining its use and category, answer whether this particular commercial requirement applies. ^[26]

The application ties the operator to acknowledgments concerning traffic-law compliance, recording capability, applicable federal requirements, a minimal-risk condition following a relevant ADS failure, registration, and required insurance or self-insurance. It also requires certification that an emergency-interaction plan has been provided to the Department of Public Safety. The authorization is therefore more specific than a general statement that the state permits autonomous driving. ^[26]

DPS publishes a first-responder interaction-plan interface and operator plans. This supplies a separate administrative artifact beyond the commercial authorization itself: responders can identify material describing how to interact with an operation. The existence of a filed plan does not prove how well personnel will perform during an actual event, but it establishes an identifiable preparation and communication mechanism. ^[29]

Enforcement has a threshold and a process

Texas distinguishes roadside enforcement from TxDMV's administration of commercial authorization. The latter can involve restriction, suspension, or revocation. Its published process focuses on operations that have resulted, or are likely to result, in serious bodily injury, and includes notification, an opportunity for correction, and review procedures. It is not an automatic fleet-wide suspension mechanism for every ordinary traffic infraction. ^{[26] [28]}

California and Texas both impose operating requirements and enforcement mechanisms, but their entry procedures, reporting structures, and administrative processes differ. Those differences do not by themselves establish which system produces better outcomes.

Arizona illustrates an earlier allocation of responsibilities

Arizona is included as a historical comparison, not because all of its relevant requirements were enacted during this review. Its statute requires a law-enforcement interaction plan and a written statement addressing federal compliance or exemption, a minimal-risk condition, state traffic-law compliance, and the relevant vehicle documentation and coverage. It also provides a recipient for citations associated with the operation and a means to halt operation when required submissions are absent. ^[30]

The state limits additional county and municipal AV-specific requirements, while preserving specified airport authority over fees and pickup or staging arrangements. State-centered regulation therefore does not mean that every location becomes operationally identical or that airport access is automatically unrestricted. The statutory allocation is incomplete without its exceptions. ^[31]

Functional overlap is not legal harmonization. Across these states, recurring functions include an identifiable operator, a basis for vehicle legality, operating conditions, fallback expectations, and a public-authority interface. The legal pathway still depends on the requirements and decision-maker in each jurisdiction.

The Next Regulatory Questions: Strategy, Competency, and Unfinished Boundaries

A national strategy is a direction, not a national license

On September 3, 2026, DOT published its National Strategy for Automated Vehicles for fiscal years 2026-2030. The strategy identifies safety, regulatory certainty, innovation, and interoperable corridors as objectives. It succeeds a history of federal AV policy documents; it is not the first time the United States has articulated a national AV policy. ^{[13] [14]}

The strategy supersedes earlier federal AV policy publications where they conflict, but it does not replace statutes, final regulations, or state operating authority. Its practical effect still depends on the legal or administrative instruments used to implement its objectives. ^[14]

Driving competency remains a different question from vehicle configuration

NHTSA's July 2026 announcement included A2SCEND, a partnership with SAE ITC described as a three-year, five-million-dollar effort to develop automated-vehicle performance standards. It also described work to update guidance on topics including emergency interaction, safety management, remote assistance, and post-crash behavior. These are development initiatives, not evidence that a comprehensive federal ADS driving-competency standard is already in force. ^[7]

Vehicle-configuration rules and driving-competency standards answer different questions. The latter would need objective evidence tied to relevant operating conditions, with enough boundary definition to avoid treating a bounded test as universal competence. Different technical architectures may supply that evidence differently.

Congressional proposals remain proposals until enacted

The retrieved Congress.gov record for H.R. 7390, the SELF DRIVE Act of 2026, records introduction on February 5 and a February 10 subcommittee action forwarding it to the full committee. It does not record enactment. The bill therefore cannot be used here as an existing nationwide AV statute or as authority for permissions that a future law might change. ^[15]

Remaining boundaries are not all the same kind of gap

Important gaps remain distinct: federal driving-competency standards are still under development; state operating processes are not harmonized; passenger-service rules address different questions from vehicle rules; and new reporting mechanisms have not yet produced long-run outcome evidence. These are different kinds of incompleteness and should not be collapsed into one maturity claim.

Accordingly, the old twelve-to-twenty-four-month market forecast has been replaced with observable institutional questions. Relevant future evidence includes completed decisions, effective requirements, published reporting, and documented enforcement outcomes. The evidence reviewed here does not support assigning a date by which the whole United States will have finished regulating autonomy.

What This Changes for AI and Robotics

Deployment becomes a maintained state

Autonomous driving still depends on sensing, edge compute, planning, and fleet operations, but continued deployment also depends on alignment between the system being operated and the legal basis for that operation. A launch is an event; authorization is a maintained condition.

When a fleet expands its operating domain or materially changes its system, the practical question is whether the new operation still fits its documented claims and authorization. Zoox's adaptable operating conditions and California's permit/reporting framework are two different mechanisms for maintaining that link after launch. ^{[6] [19] [20]}

Observability connects engineering evidence to institutions

As an engineering-governance inference, incident review becomes more useful when records can connect the vehicle, operating conditions, system configuration, and response. Event definitions, record retention, version identification, and submission responsibility therefore become interfaces between engineering operations and regulatory review.

Observability also has boundaries. Information supplied to a regulator and information released to the public are not necessarily identical. NHTSA's reporting framework recognizes confidential-business-information and personal-information issues. Public transparency therefore requires attention to what is disclosed, what is withheld, and what conclusions the available data can support, rather than assuming every required submission becomes an unrestricted public dataset. ^[10]

Fallback and recovery are different from normal driving

Texas and Arizona both connect driverless operation with a minimal-risk condition when the system cannot continue as required. That is a defined safety responsibility, not a claim that every interrupted trip will be convenient or that all road obstruction has been eliminated. ^{[26] [30]}

Stopping an unsafe continuation and restoring a disrupted service are different functions. Recovery can require communication, remote assistance, vehicle retrieval, or public-authority coordination, so human support is best described by the function it performs rather than by a generic label.

Public infrastructure is complementary, not an assumed prerequisite

The federal strategy discusses vehicle-to-everything communication and interoperable corridors as potential complements to automated transportation. It does not establish a nationwide requirement that every robotaxi depend on universal V2X infrastructure before deployment. ^[14]

The broader implication for embodied AI is that public deployment depends on more than computational capability. It also depends on intelligible interfaces with institutions that operate on different timescales: immediate emergency response, scheduled reporting, administrative review, and longer-term rulemaking. Machine action and institutional action meet at those interfaces. The present evidence makes that dependence more concrete without proving that the interfaces are complete.

How to Evaluate Future Regulatory Changes

Follow changes in authority and evidence

The most consequential future changes will come from final standards taking effect, material revisions to operating authorizations, completed reporting cycles, documented enforcement outcomes, or legislation changing agency authority. A useful reading sequence is: identify the legal instrument, the covered entity and activity, the effective date, the obligation created or removed, and the evidence that the mechanism has actually been implemented.

What would change this conclusion?

The conclusion that U.S. AV regulation became more operationally specific in 2025-2026 would weaken if the new mechanisms proved largely nominal in practice. The current record, however, includes adopted California rules and filing processes, an enforceable Texas authorization system, and a federal commercial exemption with explicit operating conditions. ^{[6] [18] [21] [26]}

The stronger claim that the framework is complete is not supported. As of September 21, California's first quarterly reporting cycle and Tesla's sworn Cybercab response were still pending; federal AV performance standards remained under development; and state operating rules remained non-uniform. New rulemaking, reporting data, or enforcement outcomes could materially change the balance described here. ^{[7] [21] [34]}

Conclusion: governable deployment is not the same as completed autonomy

The most defensible interpretation of 2025-2026 is that several U.S. institutions made the conditions of continuing driverless operation more explicit. Federal work addressed particular vehicle-standard mismatches and exemption oversight. California specified additional operating interfaces and reporting processes. Texas made commercial driverless authorization enforceable. Earlier arrangements, including Arizona's, show that identifiable responsibility was already part of the institutional foundation. ^{[4] [5] [6] [19] [26] [30]}

The reusable structural proposition is compact: machine capability becomes deployable public service only when it sits inside a legally defined operation; that operation needs accountable organizations and observable events, while failures and changed conditions need a route to response and revision. A missing link can constrain deployment even when another part of the system performs well.

Technical comparisons among Cybercab, Waymo, and Uber require a separate analysis of vehicle intelligence, sensing, fleet operations, and distribution models. Those questions are deliberately separated from the regulatory framework examined here.

Sources

Primary legal and administrative sources are listed below. Dates distinguish historical baselines, proposals, adopted rules, implementation documents, investigations, and final actions. Analytical models in the article are Hi K Robot interpretations rather than official legal classifications.

1. NHTSA / Federal Register. Occupant Protection for Vehicles With Automated Driving Systems. Published 2022-03-30 at 87 FR 18560; effective 2022-09-26. Historical final-rule baseline.

2. U.S. Department of Transportation. Automated Vehicle Framework announcement. 2025-04-24. Policy framework; historical baseline.

3. NHTSA. Domestic automated-vehicle demonstration exemption for Zoox. 2025-08-06. Demonstration exemption, not the later commercial grant.

4. U.S. Department of Transportation / Federal Register. FMVSS No. 102: transmission shift-position display requirements for ADS-equipped vehicles. 2026-03-16. Notice of proposed rulemaking; document 2026-05024.

5. NHTSA. Rulemaking to update brake-pedal requirements for automated vehicles. 2026-06-25. Proposed FMVSS No. 135 changes.

6. NHTSA / Federal Register. Zoox: grant of temporary exemption from portions of eight Federal Motor Vehicle Safety Standards. 2026-07-31. Grant; 91 FR 48494-48507; effective through 2028-07-31.

7. NHTSA. Automated-vehicle actions: Zoox commercial exemption, A2SCEND, and guidance initiatives. 2026-07-30. Mixed instruments: grant, research partnership, rule and guidance actions.

8. NHTSA. Proposed ADS-equipped Vehicle Safety, Transparency, and Evaluation Program (AV STEP). 2024-12-20 announcement; 2025-01-15 proposal. Historical proposal; subsequently withdrawn.

9. NHTSA / Federal Register. AV STEP: withdrawal of notice of proposed rulemaking. 2026-06-26. Withdrawal; document 2026-12980.

10. NHTSA. Standing General Order on Crash Reporting: requirements and data limitations. Third amendment effective 2025-06-16; retrieved 2026-09-17. Current reporting framework and methodological cautions.

11. NHTSA. Third Amended Standing General Order 2021-01. 2025-04-24; effective 2025-06-16. Binding order for identified reporting entities.

12. NHTSA. AQ26002 opening resume: Tesla Cybercab FMVSS Certification. Audit Query opened 2026-09-03; publicly announced 2026-09-04. Open investigation; not a final compliance determination.

13. U.S. Department of Transportation. National Automated Vehicle Strategy announcement. 2026-09-03. Federal policy strategy, not a nationwide operating license.

14. U.S. Department of Transportation. National Strategy for Automated Vehicles, FY 2026-2030. 2026-09-03. Strategy document.

15. Congress.gov. H.R. 7390, SELF DRIVE Act of 2026: all actions. Introduced 2026-02-05; record retrieved 2026-09-17. Legislative proposal; not enacted in the retrieved record.

16. California Department of Motor Vehicles. New autonomous-vehicle regulations: oversight, enforcement, trucks and transit. 2026-04-28. Adoption announcement; operative detail in adopted texts.

17. California Department of Motor Vehicles. Autonomous Vehicle Milestones. Retrieved 2026-09-17. Official regulatory chronology.

18. California Department of Motor Vehicles. Order to Adopt Article 3.7: testing of autonomous vehicles. 2026 adopted text. Final regulatory text; definitions, testing and remote-operation requirements.

19. California Department of Motor Vehicles. Order to Adopt Article 3.8: deployment of autonomous vehicles. 2026 adopted text. Final regulatory text; deployment, safety case, emergency interaction and enforcement.

20. California Department of Motor Vehicles. Autonomous Vehicles Program Permit Resources. Retrieved 2026-09-17. Current permit and reporting templates.

21. California Department of Motor Vehicles. AVIM 2026-001 A: reporting requirements. 2026 implementation memorandum. Permit-specific reporting schedules; first quarterly submissions due 2026-09-30.

22. California Department of Motor Vehicles. AVIM 2026-002: Notice of Autonomous Vehicle Noncompliance. Effective 2026-07-01. Implementation memorandum.

23. California Legislature. AB 1777, chaptered text. 2024-09-27 enactment; relevant implementation in 2026. Enacted statute; distinguish enactment from operative requirements.

24. California Public Utilities Commission. Autonomous Vehicle Passenger Service Programs. Retrieved 2026-09-17; programs established in 2018 and 2020. Current pilot and deployment framework with historical decisions.

25. California Public Utilities Commission. New rulemaking on autonomous-vehicle passenger service. 2025-08-28. Proceeding scope; not blanket approval of the service models discussed.

26. Texas Department of Motor Vehicles. Automated Vehicle Program. Rules effective 2026-02-27; authorization required from 2026-05-28. Current program requirements and enforcement process.

27. Texas Department of Motor Vehicles. Commercial driverless vehicle authorization application opening. 2026-04-21. Implementation announcement.

28. Texas Department of Motor Vehicles. Commercial automated-vehicle authorization enforcement and public complaints. 2026-06-02. Enforcement implementation announcement.

29. Texas Department of Public Safety. Connected Autonomous Vehicles: first-responder interaction plans. Retrieved 2026-09-17. Official filing interface and published operator plans.

30. Arizona Legislature. Arizona Revised Statutes, section 28-9702. Current text retrieved 2026-09-17. Existing driverless-operation requirements; historical comparison, not a new 2026 law.

31. Arizona Legislature. Arizona Revised Statutes, section 28-9706. Current text retrieved 2026-09-17. State preemption with specified airport authority.

32. NHTSA. Interpretation 18065.nhf: manufacturer self-certification and reasonable care. 1998 interpretation; historical explanation of the statutory model. Historical legal interpretation, read alongside the current 2026 enforcement announcement.

33. NHTSA. Part 555 letter to manufacturers: exemption process and operational oversight. 2025-06-13. Agency process guidance; predecessor to the 2026 grant.

34. NHTSA, Office of the Chief Counsel. Special Order Directed to Tesla, Inc., AQ26002. Dated 2026-09-10; sworn response due 2026-09-30. Twenty-one compulsory information requests; the order is not a final noncompliance finding.

This article provides general research and analysis, not legal advice or a compliance determination for any vehicle, operator, or jurisdiction. Applicable requirements depend on the legal instrument, vehicle, activity, location, and operative date. Company examples do not constitute endorsements or findings of fault. Later amendments, decisions, and implementation records may change the analysis.

Related K Robot Research

Relationship direction: Robotaxis and Autonomous Mobility → linked research. Canonical Relationship semantics are shown for active Nodes; contextual research is identified separately.

• From Unimate to Optimus and Unitree: How the Robotic Body Became Embodied Intelligence — depends on
• AI Civilization and Sovereign Divergence: The U.S.–China Structural Divide — forms a feedback loop with
• From Tiananmen to AI Governance: How June 4, 1989 Shaped China's AI Civilization — Comparative Reference — Governance Context

Reproduction is permitted with attribution to Hi K Robot (https://www.hikrobot.com).