Why Some Robotaxi Programmes Keep a Safety Driver and Others Don't

The seat isn't the tell. A person sitting behind the wheel doesn't necessarily mean the vehicle can't drive itself; it usually means the company hasn't applied for, or been granted, the specific permit category that allows operation without one.

Most US states that regulate autonomous vehicles split their permit system into at least two tiers. California, for instance, separates an autonomous-vehicle testing permit from a driverless deployment permit, both issued through its DMV, and other states use similar two-tier structures under different names. A testing permit typically allows a company to operate on public roads, often with a safety driver on board, while it collects data. A deployment or driverless permit, a separate and generally harder-to-get authorization, allows operation without that person present.

A program can sit under a testing permit for years, accumulating real mileage and generating the disengagement data regulators use to evaluate whether it's ready for the next tier, without ever formally applying for driverless authority. Some companies have done exactly that, either by choice or because the driverless application process in a given state has its own separate timeline and evidentiary bar.

The safety driver and the remote operator are also not the same role, and conflating them misses a meaningful distinction. A safety driver sits inside the vehicle with the ability to take control directly, typically holding specific training and often a commercial license tied to the testing program. A remote assistance operator, by contrast, works from a monitoring center and generally cannot take over steering and braking directly the way an in-vehicle driver can; the role is closer to answering the vehicle's request for guidance in an ambiguous situation than physically driving it. A fleet that has moved past requiring an in-vehicle safety driver can still rely on remote assistance, and the presence of that remote layer says nothing about whether the vehicle holds driverless authority in the DMV sense.

Some companies have also kept a safety driver in a market voluntarily, even after clearing the regulatory bar to remove one, phasing the person out gradually rather than all at once once a driverless permit is granted. That's a caution-driven business decision layered on top of the regulatory floor rather than a sign the permit hasn't actually been issued, and it's one more reason the seat itself is an unreliable signal of what authority a company actually holds in a given state.

There's also a jurisdictional split above the state permit layer worth noting. Federal regulation, through the National Highway Traffic Safety Administration, governs vehicle equipment and design standards, including questions like whether a vehicle must have a steering wheel or manual controls at all, while state permitting governs whether and how a vehicle can actually be operated on public roads. A vehicle can clear the federal equipment bar and still be waiting on a particular state's operational permit, or the reverse, which adds a second axis of approval separate from the testing-versus-deployment distinction within a single state.

The mistake is reading a safety driver as proof the underlying technology can't yet handle driverless operation elsewhere. It's a common one, and it's usually wrong. A company might run driverless in one state under a granted deployment permit and keep a safety driver in another where it simply hasn't applied for that authority yet, or where the state's process is slower.

This site's disengagement reports page tracks the mileage and incident data companies file under their testing permits, and the deployment tracker tracks which companies hold driverless authority in which states, which is the more useful question if the goal is figuring out where a company can actually operate without a person on board.