Autonomous vehicles could save lives, clean up cities and bring new mobility to people who can’t drive.
But developing autonomous technology is a messy business — as the sizzle reels from engineers touting their latest devices won’t tell you. At Disrupt, Waymo co-CEO Tekedra Mawakana will blast through the hype and tell us what is really required to run driverless cars safely, reliably and at scale.
What ‘safety at scale’ actually means for robotaxis
“Safety” tends to get the press-release adjective treatment. For a commercial robotaxi operator, it’s an engineering discipline with hard boundaries: a well-defined operating design domain, continuous scenario testing, independent oversight and transparent post-incident analysis. U.S. roads still witness 40,000-plus deaths annually, federal data shows, and that human baseline becomes the context for every AV claim.
Waymo is doing this with a focus on geofenced driverless service and a layered safety case built around redundancy between the types of sensors, prediction models and risk-averse planning. Company analyses and peer-reviewed collaborations have indicated lower rates of injury-causing collisions with human drivers on the same roads, though regulators are right to be skeptical and still scrutinize real-world performance. And federal safety investigators have pointed out in public that there have been other incidents connected to robotaxis, citing reports on Waymo vehicles and noting that even mature systems can err and must be watched closely.
The recent headline fetishization service-wide — from blocking intersections to vehicle-cyclist collisions — demonstrates the distinction between a demo and a reliable public good. Expect Mawakana to couch progress in terms of measured exposure: miles driven without a human at the wheel, and safety-critical events per million miles, not to mention solid root-cause fixes when things go wrong.
Regulation, trust, and transparency in autonomous mobility
Autonomy scaling is as much a governance project as it is a software project. In California, state regulators — namely the Public Utilities Commission and Department of Motor Vehicles — largely decide what is acceptable when it comes to commercial driverless service, even as local officials insist on guardrails around traffic flow and emergency access. Arizona has adopted a far more hands-off approach, making it an ideal testing ground for continued self-driving operations.
Public trust is established through candor, not perfection. Table stakes at this point are publishing safety methodologies, allowing research access independent of the company, releasing meaningful incident summaries, and adopting standardized metrics. Groups like NHTSA, the National Transportation Safety Board and the Insurance Institute for Highway Safety have called for more uniform reporting; advocacy organizations including PAVE have also advocated for plain-language education. Mawakana’s themes have been consistent: meeting communities where they are, and asking them not to take the technology on faith.
Operations: the less sexy, essential work of scaling
Outside algorithms, the hard problems are operational. Operating a network without drivers would include:
- Cleaning, charging, and maintaining vehicles
- Monitoring fleets with remote specialists who could offer advice during rare edge cases
- Continually updating the maps and models as cities change
The goal is to drive the ratio of remote support well below one-to-one and maintain high vehicle utilization without putting prudence at risk.
Waymo’s expansion through metro Phoenix and parts of the Bay Area, with early service in Los Angeles, has also forced the company to focus less on novelty and more on repeatability. That includes how it interfaces with ride-hailing marketplaces, where to locate pickup zones that don’t block traffic and fine-tuning service for peak-hour ebbs and flows or other special events. Unit economics get better when empty rebalancing miles drop, charging cycles synchronize with demand and software updates allow for wider weather and road coverage without jacking up risk.
Competition and the Level 2–4 split in autonomy
The market generally confuses supervised driver-assistance with full autonomy. Level 2 systems — which some automakers market as “full self-driving” although they most certainly are not — still need a human to pay attention and are still the responsibility of the driver in legal terms. Waymo’s service runs at Level 4: Nobody will sit in the front seat, within some constrained domain, and it is the system itself that takes on the driving task.
Disengagement counts become newsworthy but are a poor measure of safety; policy experts at RAND and academic researchers have long warned that the figures vary between companies and don’t correlate neatly with crash risk. More important are normalized safety metrics, raw operational availability in difficult cases and detecting, classifying, and reacting to hazards as soon as possible. Mawakana will probably say that differentiation will be built as honest-to-goodness rider experiences and safe operations take place, not by the biggest advertising dollars.
Metrics that matter for autonomy’s next phase
If the next chapter is on mainstreaming driverless mobility, we need to be scoring cooperation and real transportation value. Look for four signal measurements:
- Injury crashes per million driverless miles in the prescribed area
- Average rider wait times and completion rates during times of peak demand
- The percentage of rides completed with no remote intervention
- Cost per driverless mile headed toward proximity to what’s earned in ride-hailing
Honest conversation beats hype. Increasingly, the road toward autonomy is painstaking, audited and operationally intense — and that’s precisely why it’s starting to work in a handful of cities. Mawakana’s emphasis on safety, service quality and community trust is less glamorous than a concept car, but it’s how robotaxis transition from novelty pilot programs to public infrastructure.