For years, autonomous vehicles were perpetually five years away. The confident predictions of the mid-2010s — full self-driving by 2020, robotaxis everywhere by 2021 — collapsed against the reality that driving is astonishingly complex and that rare edge cases are common over millions of miles. By the mid-2020s, the field had both matured and sobered. And then, in a handful of cities, something real began to happen.
Waymo: The Clear Leader
Waymo, the Alphabet subsidiary descended from Google’s self-driving project, is the most advanced commercial robotaxi operator in the world. By 2025 it was operating fully driverless commercial service in Phoenix, San Francisco, Los Angeles, Austin, and expanding into Atlanta and Miami. It surpassed 150,000 paid rides per week and continued to grow, clocking millions of autonomous miles. Earlier, in 2023, it had crossed 100,000 weekly rides.
Waymo’s technical approach relies heavily on lidar, radar, and cameras, combined with high-definition mapping and extensive simulation. Its vehicles — the Jaguar I-PACE and, more recently, the Hyundai Ioniq 5 — operate in defined geographic areas (geofences) that have been carefully mapped. This constraint is a deliberate trade-off: it limits the operating area but increases reliability.
Waymo’s safety record, while not perfect, has been strong. Its published data and independent analyses have suggested its vehicles are involved in fewer injury-causing crashes per mile than human drivers in comparable settings, though comparisons are methodologically difficult. The company has also navigated the regulatory patchwork state by state, working with California’s DMV and CPUC, Arizona, Texas, and others.
Cruise: The Cautionary Tale
Cruise, backed by General Motors, was Waymo’s main U.S. competitor — until it collapsed spectacularly. In October 2023, a Cruise robotaxi in San Francisco struck and dragged a pedestrian who had been thrown into its path by another vehicle. Cruise’s initial communications mishandled the incident, and California regulators suspended its permits. A cascade of problems followed, including revelations about the company’s handling of the event, safety concerns, and leadership turmoil. Cruise paused all operations, and by late 2024, GM announced it would stop funding the robotaxi program and fold Cruise’s technology into its own driver-assistance efforts.
The Cruise episode was a case study in how quickly public trust and regulatory tolerance can evaporate. It also demonstrated that a single serious incident, poorly handled, can undo years of investment.
Tesla’s Bet
Tesla has taken a fundamentally different approach, relying on cameras alone (no lidar or radar) and a vision-based neural network, with the ambition of generalising to any road without geofencing. Its “Full Self-Driving” (Supervised) product, despite the name, requires an attentive driver and is not autonomous. In June 2025, Tesla launched a limited robotaxi service in Austin, Texas, with a small fleet and safety monitors, later expanding its service area and gradually removing front-seat supervision.
Tesla’s approach is high-reward and high-risk. If vision-only autonomy works at scale, it would be far cheaper and more generalisable. But Tesla has faced investigations from the National Highway Traffic Safety Administration over crashes involving its driver-assistance features, and its timelines have repeatedly slipped.
China’s Robotaxi Scene
China has become a major autonomous driving laboratory. Baidu’s Apollo Go operates robotaxis in Wuhan, Beijing, Shenzhen, and other cities, with large fleets and ambitious expansion plans. WeRide, Pony.ai, and AutoX are among other players. China’s dense urban environments, supportive government policy, and large data volumes have enabled rapid deployment. Regulatory frameworks, including pilot zones and license issuance, have given Chinese operators more latitude than their U.S. counterparts in some respects.
The Regulatory Landscape
Autonomous vehicle regulation remains fragmented. In the U.S., no federal law governs autonomous vehicles comprehensively; states set their own rules, creating a patchwork. NHTSA requires crash reporting for vehicles with automated driving systems, which has produced valuable — if hard to interpret — public data. In 2024, NHTSA opened an investigation into Waymo after a series of minor incidents, underscoring that even the leader faces scrutiny.
In Canada, autonomous vehicle testing is governed by provincial rules. Ontario permits testing under strict conditions, and companies have run pilots, but no province has authorised commercial driverless service at scale. Canada’s long, snowy winters present a genuine technical challenge that many autonomous systems have not fully solved.
The Hard Problems That Remain
Despite progress, autonomous vehicles still struggle with:
- Adverse weather: Heavy rain, snow, and fog degrade sensor performance.
- Unusual situations: Construction zones, emergency scenes, and human gestures that require interpretation.
- Cost: Lidar and compute hardware remain expensive, though costs are falling fast.
- Scaling beyond geofences: Expanding to new cities requires extensive mapping and validation.
The Technical Divide: Two Philosophies
The split between Waymo and Tesla is not merely commercial; it reflects a genuine engineering disagreement. Waymo’s “sensor fusion” approach combines lidar, radar, and cameras, giving the system multiple redundant ways to perceive the world. Lidar provides precise distance measurements and works in conditions where cameras struggle. Tesla’s “vision-only” approach argues that humans drive with two cameras (eyes) and a brain, so with enough data and a capable neural network, cameras alone can suffice — and the resulting system is far cheaper and easier to scale. Each side has plausible arguments. Waymo’s approach is more robust today; Tesla’s would be more scalable and affordable if it works. The market will ultimately judge.
The Mapping Question
Waymo’s vehicles rely on highly detailed, pre-built maps of their operating areas, which encode lane geometry, signage, and other features. This is a major reason its rollout is city-by-city and slow. Building and maintaining such maps is labour-intensive. Tesla’s approach attempts to drive on roads it has never mapped, using the neural network to interpret them in real time — a much harder perception problem, but one that would generalise far more easily. The mapping question is central to the economics of scaling an autonomous network.
Insurance, Liability, and the Legal Vacuum
When a robotaxi crashes, who is liable — the manufacturer, the software provider, the operator, or the remote monitor? Existing liability regimes assume a human driver. Autonomous vehicles strain those assumptions, and clarity is emerging slowly through legislation, litigation, and the practices of insurers. In the U.S., NHTSA’s mandatory crash reporting for automated systems creates a data trail, but no comprehensive liability framework. Until responsibility is clearly allocated, insurers will price uncertainty into premiums, and operators will be cautious about scaling.
Public Trust as a Bottleneck
Autonomy’s progress depends as much on public acceptance as on technology. Surveys consistently show substantial unease about riding in driverless cars, particularly after high-profile incidents. That unease translates into political pressure, which translates into regulatory caution, which slows deployment. Waymo’s careful, safety-monitor-supervised rollout is a response to this dynamic: build trust incrementally. The lesson from Cruise’s collapse is that a single mishandled incident can set the industry back years. Trust, once lost, is expensive to regain.
The Canadian Road Ahead
Canada presents both opportunity and challenge for autonomy. Its dense, well-mapped cities like Toronto and Vancouver are plausible early markets, and its research base is strong. But long winters, snow-covered lanes, and obscured road markings confound systems trained in sunnier climes. Provincial regulatory frameworks remain permissive but limited, and no province has authorised commercial driverless service at scale. Canada is likely to be a follower rather than a leader in deployment, even as it remains a source of talent and research.
The Employment Question
Autonomous vehicles raise a labour question that is often subsumed by the safety debate. Millions of people worldwide drive for a living: taxi and ride-hail drivers, delivery drivers, long-haul truckers, and bus operators. Robotaxis do not replace all of these — many tasks remain unsuited to automation — but they do threaten jobs in the most automatable segments. The transition will be uneven, hitting some cities and categories first. Whether the displaced workers find new roles, and whether policy eases the transition, is a question of political economy, not engineering. The history of automation suggests that new jobs often emerge, but also that the adjustment is painful and unevenly distributed, and that the workers who lose out are rarely the ones who capture the gains.
Conclusion
Autonomous driving in 2025 is best described as real but uneven. Waymo has demonstrated that driverless robotaxis can operate safely and commercially at meaningful scale in favourable environments. Competitors have failed spectacularly or lagged. The technology has moved from hype to incremental deployment. The question is no longer whether autonomous vehicles will exist — they do — but how fast they can expand, how cheaply they can operate, and whether they can earn and keep public trust. On that front, the cruise from hype to reality has been bumpy, and the road ahead remains long.
Note on safety culture: The autonomous vehicle industry has learned, at great cost, that safety is not a feature but a culture. The companies that succeed at scale will be those that invest in rigorous, transparent safety processes, independent oversight, and honest communication about limitations — not those that promise the fastest timeline. Waymo’s cautious expansion and Tesla’s iterative deployment represent two different philosophies of safety, and the market will ultimately decide which approach earns the public trust that is autonomy’s scarcest resource.
