First Tesla Robotaxi Rides In California May Risk DMV Shutdown

Forbes

2 days ago

Tesla has begin limited operation of a ride service in the San Francisco Bay Area. The first video by a passenger shows that it is running a similar software system to the one used in Austin TX, with a safety driver, in this case located behind the wheel rather than in the passenger seat. While the saftey driver keeps hands on the wheel, as recommended with Tesla's consumer version of FSD, in this ride they do not appear to control the vehicle, and it drives the trips, including pick-up and drop-off, without input from the Tesla employee.
The problem is, this sort of more fully automated ride may run afoul of more subtle complexities in California self-driving vehicle regulations which led to the DMV shutting down testing by Uber's self-driving research unit ATG several years ago. Tesla is using a carve-out in the California laws which state that they do not cover 'driver assist' tools, sometimes referred to as 'Level 2' in specifications from NHTSA and the Society of Automotive Engineers.
The California regulations require a large set of permits from both the DMV and the public utilities commission to operate a taxi service based on self-driving vehicles. Seven different permits are needed, and Tesla has only 2 and has not applied for the others. By declaring to the DMV that this is not a self-driving car, but rather a driver assist car that requires a human driver behind the wheel and in control, they hope to bypass the need for those permits.
The line, however, between Tesla FSD, which is indeed correctly sold as a driver-assist system and this 'robotaxi' version is challenging. Just when does a system switch from being driver-assist to prototype self-driving? For now, the DMV is accepting that and stated:
The Regulations and Driver Assist
In 2011-12, I participated in the drafting of the first drafting of the nation's first laws regulating self-driving in Nevada and California. The first laws only enabled testing, and were prompted by Google, the only company trying to do tests. Representatives from big automakers quickly joined the discussions, and they were concerned that these regulations might interfere with some of the systems they sold, such as adaptive cruise controls, and lanekeeping systems, which are known as 'driver assist' tools because a human driver is responsible for the vehicle, and the system only assists. They got the carve-out they wanted.
In 2016, Uber was developing self-driving at its ill-fated 'ATG' division. The head of the division, Anthony Levandowski, who had represented Google in the drafting of these laws, began testing their vehicles on California roads. He declared that because the vehicles had a human safety driver on board, they were driver assist, and Uber didn't need a self-driving testing permit.
The DMV would have none of it, and threatened Uber with pulling its vehicles from the roads, cancelling their licence plates. Uber complied and got the testing permits. Later, Uber ATG would have a fatal crash. It shut down operations and the team was purchased by Aurora. (Aurora just announced this week that they have begun 'driverless' trucking at night in Texas, though also with an employee behind the wheel.)
The DMV has not had enough time to look at the new service that Tesla has deployed. The Tesla robotaxi stack definitely tries to perform the complete robotaxi task, including pick-up and drop-off. It is not ready from a safety standpoint. Other data suggests the Tesla FSD system needs human intervention around every 400 miles, Tesla has said they now have reached near 10,000 miles, but their operations in Austin suggest otherwise. Either way, to make a working robotaxi requires needing a serious intervention every million miles or so to meet Musk's stated goal of 'much safer than a human' and so Tesla still has very far to go and the safety driver is needed.
At the same time, Uber ATG was very, very, very far behind this quality level. At the time of their fatality they needed safety driver takeover about every 15 miles. Because their safety driver disregarded her job and watched a video instead of the road, the vehicle struck and killed a pedestrian. Uber ATG never took passengers, so their vehicle also was not capable of doing the things like summoning, pick-up and drop-off that the Tesla vehicle does.
It seems very unlikely that an analysis of the Tesla Robotaxi system in comparison to the Uber ATG system would class the Tesla system as less of a prototype self-driving system. Except for one strange irony.
DMV Lawsuit Over FSD Name
The DMV is currently in court suing to remove Tesla from the roads in general for deceptive labeling, but in the opposite direction. Tesla calls its consumer product 'Full Self Driving (Supervised)' and formerly called it 'Beta.' The DMV has been declaring that Tesla's system is not self-driving (and indeed it isn't) and that they should not be using a name that suggests it is.
Tesla robotaxi isn't self-driving either, but like Uber ATG's system, it definitely is intended to be. Uber ATG was made to get the permits because they were trying to build a robotaxi, even though it wasn't ready yet. Tesla is very explicitly calling their system a robotaxi, though it also isn't ready yet. The DMV will have to make a decision and possibly alter its policies.
Product Quality
At present the service seems very limited. The influencer who got the early ride above got the same car every time he asked for a ride, and appeared to be followed by a Cybertruck chase car, so it was carefully monitored. However, there's no reason Tesla can't put this into operation with a safety driver. Indeed, it's no surprise that Tesla could immediately allow a larger service area than Waymo does for their actual self-driving robotaxi service. Tesla FSD with a supervisor is reasonably safe over most roads in the USA. Other than logistic costs they could offer a service anywhere, though of course it costs as much as a limo service to operate and so is not commercially interesting.
You can, and other companies have, offered test robotaxi services with safety drivers even though the robotaxi software still needs 100x or 1,000x improvement in quality in order to work. While Perhaps it only needs a 2x improvement and is thus 'almost ready' it is not easy for outsiders to judge this quality, you need statistics over large nubmers of miles, which Tesla does not release.
The robotaxi system, which has been seen in Austin, has added impressive capabilities above the point to point driving abilities that Tesla FSD has shown for some time. Most notably it is doing pick-up and drop-off on aribitrary curbsides, which took many teams some effort to develop, though again, they made it work without a safety driver, which is vastly harder. In pondering why Tesla has released this service, this may be the main reason--it already has been doing lots of testing of the FSD driving system, including in the Bay Area (the rider's route included Tesla HQ after all.) It is the PuDo (Pick-up/Drop-off) which is new and needs testing.