I suppose on some level it isn’t exactly surprising that John Krafcik, the ex-CEO of Waymo (2015-2021) doesn’t think Tesla’s Cybercab will have what it takes to be an actual, functional self-driving taxi. Waymo currently operates robotaxi services in Phoenix, San Francisco, Los Angeles, and Austin while Elon Musk, the little-known CEO of Tesla, has been promising that everyone who has a Tesla will be able to make money from their Teslas getting side jobs as robotaxis for about a decade now, has yet to make that actually happen. So it’s not too surprising to find that the former head of one of the companies actually operating robotaxis may have things to say about Tesla’s very public and hyped plans to get into the same business.
Krafcik gave an interview to Germany’s Manager Magazine earlier today, where he did a pretty comprehensive job of explaining why he felt that Tesla is nowhere near ready to start a robotaxi service, based on their current level of automated driving hardware and software.
Perhaps most obviously, Krafcik notes that Tesla’s cars, including the new Cybercab concept, which was designed specifically to be a self-driving robotic taxi, lack the type of hardware Krafcik believes is needed for an effective and safe robotaxi, including redundant sensors, high-resolution cameras, and equipment to clean cameras and sensors from grime, ice, and anything else that may impair their use.
In fact, Krafcik seems to have nothing but issues with Tesla’s design of their robotaxi, the Cybercab, saying of the prototypes:
If a company were serious about building a safe robotaxi business, the robotaxi wouldn’t look anything like this prototype. A serious robotaxi would demonstrate the primacy of safety; the manufacturer would place sensors in optimal positions—on the roof, as well as on the sides and corners of the vehicle. These sensors would also have cleaning and drying functions—windshield wipers, compressed air nozzles, and so on. A serious robotaxi also wouldn’t have a low-slung coupe body design. This design makes it difficult for people to easily get in and out; not everyone will be able to use these robotaxi vehicles comfortably.
We’ve wondered about the design of the Cybercab as well, though more about the logic of a two-seater doing taxi duty, logic that, frankly, eludes me.
Krafcik’s criticism and skepticism of Tesla’s robotaxi service, scheduled to be launched in June, is pretty dismissive, with him noting that
“There are many ways to fake a robotaxi service,”
… and implying that this is, in fact, what Tesla may end up doing. Krafcik mentioned teleoperation, cars that lead or follow, and extremely limited operating domains as ways Tesla could “fake” a robotaxi service, but he also suggested that maybe it just wouldn’t happen at all, and be another missed milestone from Musk.
I think the most surprising thing revealed in the interview had less to do with Tesla and their chances at deploying a robotaxi service by June, and more about the challenges of automated driving itself. Counter to most people’s estimations, Krafcik believes that highway driving will actually be the bigger challenge for automated vehicles.
While driving in urban, high-density areas seems like the greater challenge for a self-driving vehicle, with much more complex terrain, higher traffic density, the presence of pedestrians and animals and all manner of other obstacles and distractions, Krafcik states that the increased speeds and presence of many large vehicles like big rig trucks actually makes for a more challenging self-driving environment.
As far as the lack of “edge cases” – a term used in the self-driving world to mean pretty much any normal thing you may encounter when driving that isn’t strictly part of the expected set of elements – on highways, which most people assume would less frequent, Krafcik has thoughts on that, too, in the context of the possibility of automated trucks on highways:
Almost all of the challenging circumstances and vulnerable road users found in cities also exist on highways—only less frequently. We’ve seen cyclists, scooter riders, and pedestrians on American highways. The rarity doesn’t make things easier—it makes them more difficult. You can’t ignore these extremely rare events; you have to solve them robustly, even if the speeds are much higher and the stopping distances are much longer. This means that the sensing, perception, behavior prediction, and path planning aspects are much more demanding for autonomous trucks than for slower-moving robotaxis in the city.
The entire interview is interesting, certainly because Krafcik should understand the challenges of operating a fleet of self-driving cabs better than anyone.
Will Musk and Tesla manage to start everything in June and make Krafcik look like a fool? I suppose it’s possible, though I wouldn’t counsel anyone to do any breath-holding.
“Man thinks pathological liar who has been lying about topic for over a decade, is lying about that same topic.”
A little too wordy, but there’s your headline.