We don’t tell the car what to do: a Wayve robotaxi ride and what it means for cities

• TL;DR
• Wayve’s end-to-end AI approach (a single neural model that learns to turn sensor inputs into driving actions) has logged millions of miles and is testing in hundreds of cities, but London’s chaotic streets remain a high bar for safety and scale.
• Robotaxi deployments raise hard questions about safety, remote operators, costs and the future of urban mobility—policy choices now will lock in city priorities for decades.
• C-suite and city leaders should run tightly instrumented pilots, demand transparency on remote assistance and incident reporting, and pair any AV rollout with strong workforce transition and curb-management plans.

King’s Cross: a stress test for autonomy

King’s Cross is not gentle on software. Narrow lanes, delivery vans double‑parked, cyclists appearing from alleys and an endless fanfare of roadworks make it one of the toughest environments any autonomous vehicle (AV) can face. A Wayve‑modified Ford Mustang moved through that tangle while a human monitored, ready to intervene. The ride felt less like following instructions and more like watching a driver who had learned to read the street’s mood.

That single ride matters because it crystallises a broader shift: robotaxis are moving beyond controlled demos into the messy, rule‑bending reality of cities. The technology choices companies make—how they sense the world, how they train their models and how they handle exceptional situations—will determine whether robotaxis reduce harm or simply scale existing problems.

How Wayve drives — and how that differs from rivals

Wayve, founded in 2017 by Alex Kendall and Amar Shah, pursues an “end‑to‑end” approach: a single, large neural network trained to map raw sensors to driving actions (in short: cameras, lidar or radar → model → steering/throttle decisions). The company reports more than 7 million autonomous miles and tests in 500+ cities. Backers and partners include Mercedes, Nissan, Stellantis, Microsoft, Nvidia and Uber, and the firm’s valuation has been reported in the billions.

“We don’t give the car step‑by‑step orders — it learns to read people’s body language and predict what will happen next.”

— Alex Kendall (paraphrase)

Contrast that with other players:

• Waymo: a mapping‑and‑redundancy play. Heavy use of high‑definition maps, lidar and modular software stacks designed for explicit fallbacks. Operates at scale (reports indicate roughly 450,000 rides per week in its operating areas).
• Baidu / Apollo Go: a China‑focused scale strategy with local partnerships—reported at around 250,000 rides per week.
• Tesla: camera‑first and consumer‑centric; pushing software into customer vehicles and experimenting with a two‑seat “Cybercab” concept, while its US robotaxi pilots still typically require supervised drivers.

Each architecture has tradeoffs. End‑to‑end models can generalize from varied streetscapes if trained broadly, potentially reducing reliance on detailed maps. Modular stacks emphasise redundancy and interpretability—useful when regulators want a human-readable chain of decision‑making. Sensor choices (camera vs lidar vs radar) affect cost, robustness in poor weather and the complexity of perception pipelines.

Safety record, failure modes and the human backstop

Overall safety statistics for AV fleets show improvement over time, but notable failure modes persist. Examples reported across deployments include vehicles disabled with a traffic cone, AVs used to trap a passenger, systems that froze during a citywide power outage in San Francisco, and a case where an AV blocked emergency vehicle access in Austin.

Operators are candid that automation today still needs human rescue. Waymo has told regulators it uses human “remote assistance” for difficult edge cases; public reporting has suggested Waymo runs several dozen remote assistants on duty for thousands of vehicles. That practice prompts real concerns about latency, oversight, and the security and labour implications of routing critical decisions to remote staff—sometimes based in other countries.

“Relying on remote operators abroad for AV edge cases raises safety and cybersecurity concerns and risks exporting jobs.”

— Senator Ed Markey (paraphrase)

The practical implication: until systems demonstrably handle rare or unusual situations (a mattress in the road, an impromptu street protest, an unexpected surge of cyclists), regulators and operators will need clear, audited protocols for when and how humans intervene. Those protocols must be fast, transparent and secure.

Regulation and market numbers: the UK entering the race

The UK’s Automated Vehicles Act (2024) establishes a legal path for robotaxi approvals and paves the way for pilots and commercial services in British cities. That legislative shift coincides with growing private investment and tests: Wayve’s claimed footprint, Waymo’s operating scale and Baidu’s rapid expansion all indicate a market moving from pilots to service trials.

Context matters. England has roughly 56,400 licensed taxis and 256,600 private‑hire vehicles (2024). Market trends show private‑hire numbers rising (up about 10.5% in 2023–24) while licensed taxis declined modestly. Those shifts already reshape livelihoods in the gig economy and taxi sectors—robotaxis will accelerate the pressure.

Robotaxi economics: why price still matters

For consumers, price and convenience usually beat novelty. An independent study from San Francisco found Waymo rides cost roughly 12.7% more than an Uber and about 27.3% more than a Lyft on average. Several cost drivers explain this gap:

• CapEx: sensors, compute and vehicle retrofits—lidar and high‑end compute stacks add expense.
• OpEx: remote assistance staffing, higher maintenance rates for fleets operating 24/7, and regulatory compliance costs.
• Utilization: profitability hinges on keeping vehicles busy. Empty rebalancing miles (driving without passengers) erode margins.
• Insurance and legal risk: premiums and reserves for edge cases remain a line item until liability norms settle.

That means robotaxis are not a plug‑and‑play cost win today. Lowering fares requires higher utilization, cheaper sensors/compute, or operating models that subsidise rides (municipal contracts, transit agency partnerships) while the technology matures.

Risk snapshot for executives

• Safety risk: rare incidents can damage public trust; track incident rates per 100,000 miles and transparent investigations.
• Cybersecurity: remote interfaces and teleoperation increase attack surfaces—require hardened comms and identity controls.
• Labor and political: displacement risk for drivers, potential union pushback and reputational damage.
• Regulatory: shifting rules on liability, data access and local approvals can change operating economics quickly.

Urban policy: do robotaxis lock in the wrong future?

Deploying robotaxis is not a purely technical choice; it’s a city planning decision. Prioritising curb space, creating kerbside pickup zones and incentivising vehicle miles travelled will shape congestion, equity and carbon outcomes. If cities channel scarce curb and road resources to robotaxis, they risk cementing car‑centric mobility that undermines public transit, walking and cycling.

“Drivers provide more than driving—handling lost property, assisting passengers, making route judgments—which a robotaxi can’t easily replicate.”

— Steve McNamara, Licensed Taxi Drivers’ Association (paraphrase)

Policymakers should weigh tradeoffs: robotaxis can improve access for people with mobility constraints and reduce driver risk, but they can also increase empty traffic, displace decent jobs and concentrate control of transport in a few private platforms unless counterbalanced by regulation and public investment.

For C‑suite and city leaders: practical next steps

Run pilots, but run them like experiments—with KPIs, public reporting and worker transition plans. Your checklist:

• Define safety KPIs: incidents per 100k miles, remote assists per 1,000 trips, mean time to recover from a failure.
• Mandate transparency: incident disclosures, remote assistance logs, data access for regulators and independent auditors.
• Measure economics: cost per mile, utilization rate, average trip fare vs incumbent ride‑hailing, and subsidy needs.
• Protect curbspace: time‑limited pickup/dropoff bays and dynamic pricing to disincentivise deadheading.
• Plan workforce transitions: retraining funds, redeployment programs, and partnerships with local unions and training providers.
• Harden cybersecurity: authenticated remote operator channels, encrypted telemetry and regular third‑party penetration testing.

Metrics dashboard to require from pilots

• Trip volume and utilization (trips per vehicle per day)
• Incidents per 100k miles and breakdown by severity
• Remote assists per 1,000 trips and average latency to intervene
• Customer satisfaction and accessibility service scores
• Cost per mile and subsidy per trip (if any)

Policy makers: three fast rules

• Insist on local oversight: require physical or legal presence for remote operators and clear chains of liability.
• Protect public value: set data sharing requirements (anonymised trip/incident data) and limit undue exclusivity for platform providers.
• Balance mobility investments: tie any public support for AVs to parallel investments in transit, active travel and equitable access programs.

Key takeaways

• Robotaxis are real—but not yet frictionless. Technology has progressed, but rare edge cases, operational costs and human factors still demand oversight.
• How you deploy matters as much as whether you deploy. Choices about curb management, pricing and data governance will determine if AVs relieve or worsen congestion and inequity.
• Transparency and KPIs beat marketing: demand incident data, remote‑assist logs and concrete cost models before scaling.
• Plan for people: workforce transition and clear regulatory guardrails will decide the political feasibility of any roll‑out.

Robotaxis are a hinge point, not a hammer. If leaders treat them as a gadget to be added to the status quo, cities will pay in congestion, lost jobs and stunted mobility choices. If they treat robotaxis as one tool in a broader mobility strategy—with strict pilots, public accountability and active investment in transit and active travel—robotics and AI can amplify the public good rather than capture it.