Waymo's Testing Halt Marks a Critical Turning Point for Cities to Get Ahead in t

A Pause is Not a Failure, But a Strategic Window for Cities to Reclaim Dominance

The answer is clear: this is a rare breathing and planning space for cities and regulators. Over the past decade, we have witnessed tech platforms charge ahead in transportation and accommodation under the banner of “disruptive innovation,” by the time regulations catch up, established facts and user habits are already formed, and bargaining chips are lost. The essence of Waymo’s testing pause is New York City’s refusal to hand over “sovereignty” of public roads under conditions of data black boxes and unclear safety commitments. This is not anti-technology, but a demand for a fairer, more transparent negotiation of the rules of the game. For cities worldwide watching—including Taipei, Singapore, London—this demonstrates that dominance can be contested, with the key being whether they are prepared with a public-interest-based tech governance framework.

From Uber to Waymo: Have We Learned Our Lesson?

Looking back at the explosion of ride-sharing in the 2010s, its core narrative is strikingly similar to today’s autonomous driving: reducing private vehicles, improving road use efficiency, and creating a safer environment. Yet, what did we end up with? According to New York City’s own reports, from 2013 to 2017, vehicles increased by over 100,000 due to the entry of Uber and Lyft, with speeds in core Manhattan dropping by a staggering 22%. The “solutions” promised by tech companies, in the absence of volume control and right-of-way redistribution, instead became catalysts for problems.

Autonomous driving faces the same “induced demand” trap. A vehicle without a human driver, with its operational cost structure, will encourage more “deadheading” to shorten passenger wait times, and potentially cheaper services enticing people to abandon public transit. A 2025 simulation study by MIT indicated that in a city with full autonomous driving and no regulatory measures, total vehicle miles traveled could increase by 15% to 30%.

The table below compares the rhetoric and potential pitfalls of the two waves of mobility revolution:

timeline
    title Key Milestones in Autonomous Driving Regulation and Technology Development
    section Technology Incubation Period
        2010s : Tesla Autopilot launched<br>ushering in the consumer ADAS era
        2016-2020 : Waymo in Phoenix<br>launched early rider program
    section Expansion and Regulatory Collision Period
        2023-2025 : Cruise San Francisco incident triggered<br>nationwide safety review
        2025 : Multiple cities demanded<br>mandatory safety data disclosure
        2026 Q1 : Waymo New York testing permit expired and paused<br>regulatory turning point
    section Future Key Nodes
        2026-2027 : Expected rollout of<br>unified data standard frameworks by countries
        2028+ : L4 service commercialization and scaling<br>depends on regulatory clarity

The Black Box of Safety Data: An Unverifiable Game of Trust

The autonomous driving industry has long relied on the statistic that “human drivers are the cause of 94% of accidents” to establish its safety legitimacy. However, as pointed out by the Chair of the U.S. National Transportation Safety Board (NTSB), this figure has been misused and even removed from promotional materials by its original source—the National Highway Traffic Safety Administration (NHTSA). It over-simplifies accident causes as “driver error,” ignoring systemic factors like road design, vehicle engineering, and social context.

The more severe issue is that the new data meant to replace the old narrative is equally opaque. What “near-miss” cases did Waymo generate during its months of testing in New York? How does its AI interpret police gestures or construction worker guidance at complex New York intersections? These critical “edge case” data have never been made public to the public or independent researchers. We are caught in a paradox: society is asked to trust a system claimed to be many times safer than human drivers, yet not provided with evidence sufficient for peer review.

This data black box leads to confused market perceptions. For example, Bloomberg Intelligence once cited data suggesting Tesla’s Autopilot had a superior accident rate per million miles, while a Washington Post investigation noted Waymo had a lower disengagement rate in complex urban areas. But such comparisons often rely on data with different definitions and environments, like comparing apples and oranges. The lack of a rigorous, standardized, and public accident reporting system like in aviation is the biggest obstacle to autonomous driving winning public trust.

Tesla vs. Waymo: Who Will Win the Endgame of Technology Pathways?

This is not just a competition between two companies, but a clash of two AI philosophies, two business models, and even two visions of the future world. Tesla’s “vision-only” path bets that end-to-end neural networks can learn to drive directly from camera inputs like the human brain, with advantages in low hardware cost and massive data collection scale (via a million-vehicle fleet). Waymo adopts a “multi-sensor fusion” path, combining lidar, radar, and high-definition maps, pursuing ultra-high reliability within specific geofenced areas.

Currently, this race shows an interesting “scissors gap” dynamic: Tesla leads in technology generalization capability (usable anytime, anywhere) and cost, but its absolute safety performance in complex urban areas is frequently questioned; Waymo provides a smooth experience close to human driving in deployed areas (like Phoenix), but its high hardware cost of over $200,000 per vehicle and geofencing limitations make scaling expansion difficult.

mindmap
  root(Autonomous Driving Technology Pathway Competitive Landscape)
    (Tesla (Vision-Only Faction))
      Core Advantages
        Extremely low hardware cost<br>relying solely on cameras
        Massive data scale<br>real-world million-vehicle fleet
        No geofencing restrictions<br>theoretically usable globally
      Key Challenges
        Handling edge cases<br>(e.g., strong light, anomalous objects)
        Safety validation<br>lack of公认 standards
        High regulatory approval threshold
    (Waymo (Multi-Sensor Fusion Faction))
      Core Advantages
        Multi-redundant perception<br>lidar+radar+vision
        High-definition map prior knowledge
        Proven high reliability<br>within specific areas
      Key Challenges
        High per-vehicle cost
        Slow expansion due to<br>map creation and maintenance
        Business model<br>requires high utilization to amortize costs
    (Potential Fusion or Third Path)
      Low-cost solid-state lidar<br>lowering fusion scheme threshold
      Vehicle-to-everything (V2X)<br>placing some intelligence on the road
      Open-source platforms<br>like NVIDIA DRIVE

From an industry impact perspective, if Tesla’s path succeeds, it will彻底颠覆 the essence of cars as “hardware commodities,” making them entirely software-defined mobile terminals, with profit sources shifting from one-time sales to ongoing software subscription services. Waymo’s path is more like traditional transportation services, aiming to provide safe, reliable robotaxi services in controlled environments. The success or failure of each will determine the power distribution in the automotive industry value chain over the next decade: will it lie in the software hands of automakers and tech giants, or be concentrated in a few autonomous driving service operators?

Chips and Algorithms: The Invisible Track for Taiwan’s Tech Industry

While global attention focuses on整车 applications, Taiwan’s tech industry is quietly playing an indispensable role in the underlying architecture of autonomous driving. This extends beyond TSMC’s manufacturing advantage in advanced process automotive chips to the entire ecosystem:

1. Key Sensor Components: Laser emitters for lidar, MEMS micromirrors, millimeter-wave chips for radar—all require precise semiconductor manufacturing and packaging technology.
2. Edge Computing Platforms: Autonomous driving needs to process terabytes of sensor data in real-time, driving demand for high-performance, low-power车载 AI computing chips (like NVIDIA Orin, Qualcomm Snapdragon Ride), whose manufacturing and peripheral design chains are closely linked.
3. Software and Verification Tools: As functional safety standards (like ISO 26262) become regulatory thresholds, demand for related software testing and simulation verification platforms has surged. Taiwan’s ICT industry’s experience in software engineering and system integration has the opportunity to develop specific solutions for autonomous driving.

Taiwan’s opportunity lies in avoiding the red ocean competition of整车 brands, instead focusing on becoming a “key component and solution provider for the autonomous driving era.” For example, developing traffic management systems for mixed traffic (with both autonomous and human-driven vehicles), or high-precision positioning and communication integration solutions—these are ideal arenas for Taiwan to combine its ICT and semiconductor strengths.

The Rise of Regulatory Technology: From Passive Approval to Active Shaping

Waymo’s pause event marks a paradigm shift in regulatory thinking. Past regulation was often criticized as “not understanding tech,“只能事后追赶. But now, leading cities are realizing they must use technological means to regulate technology, i.e., “Regulatory Technology” (RegTech). This means:

• Establishing Mandatory Data Exchange Interfaces: Requiring autonomous vehicles to upload key events (like system disengagements, sensor failures) in standard format to city data platforms in real-time, not just providing filtered reports after the fact.
• Developing High-Fidelity City Digital Twins: Conducting large-scale stress tests on autonomous driving algorithms in virtual environments, simulating extreme weather, special events, etc.,前置部分安全验证 before实车上路.
• Setting Dynamic Operational Permit Conditions: Instead of one-time long-term permits. For example, linking permits to specific performance metrics, such as deadheading mileage比例, contribution to public transit connectivity, or even service accessibility in disadvantaged communities.

The table below outlines possible elements of a proactive regulatory framework:

This shift in regulatory mode requires city governance teams to possess unprecedented tech literacy and data analysis capabilities. It is no longer单纯 a transportation or legal issue, but a new field spanning software engineering, data science, and public policy. For tech companies, this also means higher cooperation thresholds, but同时 brings the benefit of regulatory clarity—clear rules are always more conducive to long-term investment than unpredictable administrative干预.

Smart Infrastructure: Roads Must Also Learn to “Talk”

The ultimate vision of autonomous driving should not just be making cars smart, but making the entire transportation system smart. This involves investment in vehicle-to-everything (V2X) infrastructure. Imagine traffic lights directly sending countdown seconds to vehicles; temporary signs in construction zones being reliably read by autonomous vehicles via wireless signals; or even roads themselves sensing black ice or积水 and warning approaching vehicles.

This sounds costly, but if integrated with cities’ existing infrastructure renewal cycles (like laying fiber optics, updating traffic signal controllers) and adopting gradual deployment strategies, it is not unfeasible. For example, prioritize deployment in complex areas like commercial districts, school zones, or high-accident路段. The key is establishing open communication standards to avoid vendor lock-in. The U.S. Department of Transportation has already promoted related initiatives in this field, such as the “Intelligent Transportation Systems Strategic Plan 2025-2029.”

The启示 for Taiwan is that with our existing foundation in smart cities and ICT network construction, we can choose specific demonstration zones (like Nangang Software Park, Kaohsiung Asia Bay Area) to pilot the integration of autonomous buses and smart road infrastructure. This not only serves transportation but can also package related sensor standards, communication protocols, and data application experience into exportable solutions.

The Next Five Years: Will We See Divergence or Convergence?

Looking ahead to the autonomous driving industry post-2026, we likely will not see a single, globally consistent future scenario, but significant divergence based on regional regulatory philosophies, urban density, and public acceptance:

• “Safety-First” Closed Ecosystem Zones: Like Singapore, parts of European cities, may allow Waymo-style services to operate under strict geofencing and data monitoring, but expansion will be slow.
• “Innovation Experimentation” Open Testing Zones: Like parts of Arizona, Texas in the U.S., continue to offer宽松 environments to attract company testing, but可能伴随 public safety controversies.
• “Public Transit Integration” Hybrid Model Zones: Like Zurich, Helsinki, strictly position autonomous vehicles as “first-mile/last-mile” connectors for public transit, using policy to ensure they do not compete with rail transport.
• “Private Vehicle Upgrade” Consumer Tech Zones: This will be the main battlefield for Tesla’s path, with advanced driver-assistance features持续渗透 to individual owners, but fully driverless progress may be slower than expected.

For consumers and citizens, rather than asking “when will autonomous driving arrive,” the better question is “what kind of autonomous driving future do we want?” Is it a future where convenience for a few exacerbates social inequality and urban sprawl? Or one that strengthens public transportation, promotes equity, and enhances urban livability? The pause in testing is not an endpoint, but a starting point for this crucial conversation.