2026-06-18 — views

Physical AI Global Regulatory Map — US, EU, China and Japan Policy Divergence

US fragmentation, EU gatekeeping, China acceleration, Japan gradualism — why Waymo is in Phoenix but not Paris, and what each framework means for AV timelines.

Article 61 in the Physical AI Benchmark Series — The Regulatory Map That Explains Where AVs Can Actually Operate

The most consequential factor shaping which cities have commercial driverless robotaxis is not sensor capability, not neural-network accuracy, and not fleet economics. It is the regulatory framework of the jurisdiction in which a company wants to operate. Waymo is in Phoenix, San Francisco, Austin, and Atlanta — not Paris, Berlin, Tokyo, or Shanghai — because of regulatory architecture, not technology limitations.

Understanding the four dominant regulatory philosophies — US permissive-but-fragmented, EU type-approval gatekeeping, China government-accelerated rollout, and Japan structured gradualism — is essential for anyone trying to map how and when driverless commercial operation scales globally.

Section 1 — The US Framework: Permissive But Fragmented

The United States has no federal pre-approval requirement for AV deployment. The National Highway Traffic Safety Administration (NHTSA) sets vehicle safety standards but does not require federal sign-off before a company begins testing or deploying autonomous vehicles. This permissive federal stance created an opening that was quickly filled — unevenly — by individual states.

Dimension	Detail
Federal authority	NHTSA sets vehicle safety standards; no federal pre-approval required for AV deployment
State authority	States regulate AV testing and deployment permits; each state writes its own framework
Result	A patchwork: Waymo needs separate permits in California (CPUC), Arizona (MVD), Texas (TxDMV), and Georgia (GDOT) — each with different requirements
California	Most demanding; CPUC Autonomous Vehicle Passenger Service permits required; mandatory accident reporting; Waymo received full driverless commercial permit in 2023 after years of permitting
Texas	Most permissive; no AV-specific state permit required beyond standard vehicle registration; this is why Tesla chose Austin for its Robotaxi launch
Arizona	Early adopter; executive order removed regulatory barriers in 2015; Waymo has operated driverlessly in the Phoenix metro since 2018
Federal legislation	SELF DRIVE Act never passed; AV START Act (2023 push) also stalled; no unified federal framework exists

The practical consequence for Waymo: every new city requires a fresh regulatory engagement — months to years of permitting, mapping, and safety demonstration with that specific jurisdiction’s regulator. The Phoenix approval does not transfer to Phoenix’s neighboring city.

The practical consequence for Tesla: Texas permissiveness enabled the Austin Robotaxi launch under supervised (not driverless) operation without an AV-specific state permit. But scaling nationally still requires state-by-state regulatory work, and the states that matter most for volume — California, New York, Illinois — are not Texas.

The legislation gap matters. Multiple attempts at a federal AV framework have failed in Congress. Without federal preemption, the 50-state patchwork is the default. This creates a built-in scaling bottleneck that is structural, not temporary.

Section 2 — Europe: Type-Approval Gatekeeping

Europe takes the opposite approach from the US at the federal level: it has a centralized, harmonized type-approval system through the European Union, supplemented by UN Economic Commission for Europe (UNECE) Working Party 29 (WP.29) international standards. The result is higher regulatory coherence across member states but a significantly more demanding pre-market approval pathway.

Dimension	Detail
Framework	EU type-approval system (EU-wide harmonized); UN ECE WP.29 sets international standards including ALKS
ALKS Regulation	UN Regulation 157 — first international regulation for automated driving up to 130 km/h on motorways; approved 2021; allows eyes-off driving in limited conditions
Level 3 limit	Europe permits L3 (automated driving with driver takeover capability) up to 130 km/h on highways under Reg 157; BMW and Mercedes received the first L3 type approvals in Germany (~2023)
Level 4 commercial	No EU-wide framework for full L4 commercial robotaxi services yet; each member state must independently interpret
Germany	§1e StVG (German Road Traffic Act) pioneering law allows L4 operation in defined areas with remote operator; first L4 approvals possible in Germany before an EU-wide framework
UK (post-Brexit)	Automated Vehicles Act 2024 passed; creates a regulatory pathway for L4 commercial deployment; DLUHC and DVSA working on implementation
France, Spain, Italy	No L4 commercial driverless frameworks established
Tesla FSD in EU	Supervised FSD not yet approved under EU type-approval as a driving automation system; Tesla sells hardware capable of FSD but software capability is constrained by type-approval limits
Waymo in EU	No European operations; EU regulatory complexity combined with right-hand vs. left-hand traffic variation represents a major barrier

The EU gap is the most striking divergence from the US. Europe is the largest car market in the world by passenger car density, with exceptional road infrastructure and a wealthy, urban consumer base that would be highly receptive to robotaxi services. Yet L4 commercial driverless operation is not available in any European city and has no clear near-term pathway at the EU-wide level.

Germany’s §1e StVG is the most advanced member-state framework — it explicitly allows L4 operation in defined geographic areas with a remote operator in the loop. This could enable limited commercial robotaxi deployments in Germany before EU-wide harmonization arrives. But Germany-only approval does not scale to the EU single market.

The type-approval system matters for Tesla specifically. FSD requires EU type-approval as a driving automation system — not just hardware certification. Tesla’s vehicles ship with Full Self-Driving hardware to European buyers, but the software capability is restricted by the absence of EU type-approval for the automation functions. This is a regulatory ceiling, not a technology ceiling, and resolving it requires formal UNECE certification engagement that Tesla has not completed.

Section 3 — China: Government-Accelerated Rollout

China’s approach is categorically different from both the US and EU. The central government has designated AV leadership as a strategic national priority, embedded it within “Internet Plus,” Made in China 2025, and the broader New Energy Vehicle mandate, and is actively deploying government resources — infrastructure, testing permits, city-scale pilot programs — to accelerate domestic AV development.

Dimension	Detail
National strategy	Central government treats AV leadership as a strategic priority; top-down push through NEV mandates, smart road infrastructure investment, and national testing programs
Smart road standard	China mandating C-V2X (cellular vehicle-to-everything) RSUs on major highways; domestic AV companies required to use government-standard infrastructure
Robotaxi leaders	Baidu Apollo (Wuhan, Chongqing, Shenzhen — hundreds of vehicles); WeRide (Guangzhou); Pony.ai (Beijing, Guangzhou, Shenzhen)
Wuhan milestone	Baidu received approval for driverless (no safety driver) robotaxi operation in Wuhan in 2022 — one of the first cities globally to authorize commercial driverless rides
Data localization	All training data collected in China must remain in China; foreign companies cannot use China data in global models; creates a data-moat separation between domestic and foreign AV systems
Foreign access	Tesla operates in China but FSD has not received MIIT approval for Chinese operation; data residency requirements necessitate a separate training dataset and model (Tesla has announced plans for a China-specific FSD stack); Waymo has no China presence
National testing database	MIIT maintains a national AV testing database; accumulated over 100 million test kilometers across approved platforms by 2025 (est.)
Domestic supply chain	Government subsidizes AV development; CATL batteries, Huawei sensors, BYD chassis — a full domestic AV supply chain is being built to compete globally

The data localization rule is one of the most consequential and underappreciated structural facts in global AV competition. Chinese regulators require that training data collected within China’s borders remain in China. This means that even if Waymo or Tesla operates in China, they cannot incorporate Chinese driving data into the training pipelines that improve their global fleet. The inverse also applies: Chinese AV companies cannot freely use foreign training data. The result is two separate AV ecosystems with incompatible data moats — and China’s domestic ecosystem has the world’s largest urban driving population generating training data.

For Baidu, WeRide, and Pony.ai, this is a structural advantage at home. For Waymo and Tesla, it means China is not just a market to enter — it is a parallel technology track that requires a dedicated investment to participate in.

Section 4 — Japan: Structured Gradualism

Japan’s approach differs from all three above. It is not permissive like the US (there is no free-market experimentation), not harmonized-centralizing like the EU (the EU process is a vehicle type-approval mechanism), not state-accelerated like China. Japan’s framework is conservative, thorough, and designed around Japan’s specific demographic challenge: an aging and rapidly depopulating rural population that needs last-mile mobility solutions that conventional private car ownership cannot efficiently serve.

Dimension	Detail
Framework	Road Traffic Act amendments enable L4 operation in “specific areas” (geofenced, low-speed) since 2023
First approvals	Japanese government approved the first L4 self-driving services in 2023 — limited to rural areas with low traffic density, under 30 km/h (est.)
Focus areas	Last-mile mobility in depopulated rural communities; shuttle services on campuses and resort areas
Companies	Toyota (Woven City / WovenPlanet), Honda (Legend L3 type-approval 2021), ZMP, Tier IV (Japan-based AV stack)
Honda Legend	World’s first L3 production vehicle with type-approval; traffic jam pilot up to 50 km/h; 100 units initially leased (2021)
Philosophy	Conservative but thorough; Japan’s Ministry of Land, Infrastructure, Transport and Tourism requires extensive safety data before expanding approvals
Waymo / Tesla	No Japanese operations; right-hand traffic is an operational requirement both can technically support, but regulatory engagement is required for either to enter

Honda’s Legend is notable because it received L3 type-approval — the first production vehicle in the world to do so — before Germany’s BMW and Mercedes achieved the same milestone under UNECE Reg 157. But the Honda Legend program was limited: 100 units, leased only, not sold. Japan’s regulatory philosophy values thoroughness over speed. The government will not expand L4 approvals to urban high-density environments until the rural geofenced deployments have generated an extensive safety evidence base.

For global AV companies, Japan is a high-value potential market — the urban density of Tokyo and Osaka, high purchasing power, a cultural receptiveness to automation in service contexts — but a slow-ramp regulatory environment. The companies most likely to lead in Japan are domestic: Toyota’s Woven City project is explicitly designed to generate the safety evidence that would support future broader approvals under Japan’s evidence-based regulatory philosophy.

Section 5 — What the Regulatory Map Means for the Tesla vs Waymo Race

The regulatory map translates directly into the competitive positions of the two leading driverless candidates. Neither company’s technology advantage is cleanly convertible into global market access — both face jurisdiction-specific regulatory constraints that throttle the rate at which their technology can be monetized.

Dimension	Waymo	Tesla
US expansion	Requires CPUC / state permits per city; each city takes 1 to 3 years of permit engagement; currently operating driverlessly in 4 to 5 cities	Texas permissiveness enabled Austin launch; FSD supervised in approximately 40 US states; driverless requires state-by-state approval (zero driverless approvals as of mid-2026)
EU	No EU pathway; not operating there	FSD blocked by EU type-approval; hardware ships but capability is capped; needs formal ALKS / UNECE certification pathway
China	No China operations; data localization prevents model sharing with global fleet	FSD pending MIIT approval; separate China data stack required; unique structural challenge
Japan	No Japan operations	No Japan operations
Regulatory moat	Each city approval is hard-won competitive advantage; Waymo’s existing permits are not easily replicated by a new entrant	Supervised FSD at scale gives de facto broad presence in permissive jurisdictions even without driverless approvals; awaiting the regulatory threshold moment
Who benefits from clarity	Waymo more immediately — a clear L4 framework per jurisdiction converts its operational lead directly into market advantage	Tesla more structurally — simpler vehicle-centric regulation (vs. service-level robotaxi regulation) plays to its consumer car model

The single largest pending catalyst in global AV markets is an EU-wide L4 commercial framework. Europe has the market size, the infrastructure, and the consumer demand. If the EU establishes a harmonized L4 commercial pathway, it would be a larger addressable market opening than any single US city approval. Both Waymo and Tesla would benefit, but the EU type-approval architecture favors established vehicle manufacturers and companies willing to engage its centralized certification process.

China represents the most consequential AV market where neither Waymo nor Tesla has a functional driverless product — and where the data localization wall means that gap cannot be closed by exporting the global model. Baidu, WeRide, and Pony.ai are not just ahead in China; they are building a data moat that compounds with every driverless kilometer driven in Chinese cities.

The regulatory map is not a temporary obstacle. It is a structural feature of the physical AI deployment environment that will shape which companies reach commercial driverless scale in which markets — and on what timeline — for the next decade.

Sources: NHTSA AV policy framework — nhtsa.gov; UN Regulation 157 ALKS — unece.org; Automated Vehicles Act 2024 — bills.parliament.uk; Baidu Apollo investor relations — ir.baidu.com; Japan NPA Road Traffic Act amendments — npa.go.jp. All figures marked (est.) are estimates derived from public company materials, government publications, and industry research. They have not been independently verified and should be treated as directional. This article does not constitute investment advice.