Will self-driving cars happen soon? What if they don’t?
Late 2022 has seen many grim declarations of the future of self-driving, particularly after the decision by Ford/VW to close down Argo.AI, their self-driving startup. Other companies in the self-driving space and LIDAR space have seen market devaluations even worse than the already horrible tech space, and startups are having trouble with funding. There’s lots of doom and gloom, though in counter to that MobilEye has a great IPO and Waymo, Cruise and others have announced new milestones and improvements.
Yesterday, I outlined why the reports of the death of self-driving cars are greatly overstated. Car OEMs never wanted this to happen fast and their cold feet don’t reveal much. A lot of people got wrong impressions (particularly from car OEMs who were just guessing) of dates. Companies made predictions of when they would start having working vehicles and many presumed that meant they would be everywhere. After all, if a car company says a car is coming out in 2023, you presume that absent short supply, you can get it in 2023. Robotaxi plans have always been for limited deployment areas.
Car OEMs are not in a rush here. They don’t want their industry turned upside-down by a tech revolution. They can’t stop revolutions but they want them to be done by them, at their pace, not at the pace of companies like Google. It doesn’t take much to convince them to slow down from that pace.
More tractable problems
So while robotaxis have happened this decade, it’s in a limited way. And many have declared that we won’t see them for a decade or two. While many (including myself, of course) don’t think that’s true, it’s worth considering what might happen if they are right and the cars don’t show up. This is not an idle question — for smaller companies, without the resources of Google, Amazon, GM or Apple, a smaller, more tractable problem is absolutely the right choice.
Cars that are hard to crash
Luminar LIDARs are being sold mostly for driver assist, to make cars that crash less
Many of the car OEMs have declared that they will scale back from self-driving efforts to driver-assist or ADAS. They were doing that before this all started and plan to keep doing it, but the computing and sensor technology built in part for self-driving is going to produce cars that are much harder to crash. This is much more tractable because you don’t have to solve all the driving problems, you just have to notice when a driver is doing something wrong that can be corrected.
The good news is that this should greatly improve road safety. One might view that as “bad” news for self-driving cars, since one of the strong motivators of robocar development has been the opportunity to make cars much safer. That can indeed happen, but it’s going to happen anyway even with human drivers, at least to some degree.
Less certain is the fate of products like Tesla’s “FSD,” which today, and possibly for a long time, is a driver assist tool where the car drives but a human watches attentively. There are issues with how attentive humans are, how this gets worse the better the system gets, and whether or not it’s actually a useful relaxing product on city streets the way it can be on freeways.
As such, while ADAS will improve driver safety, it won’t make it perfect, nor likely as good as the robocars will eventually make it as they move to being superhuman at preventing accidents, including accident avoidance when not at fault.
Delivery & Trucking
Delivering cargo is a more tractable problem. You don’t have to worry about passenger safety at all, and you can solve the problem of safety for other road users by limiting speed and deliberately picking slower and longer — but safer — routes in a way passengers would not tolerate. You can make your vehicle smaller and lighter.
In a more extreme case, there are sidewalk delivery robots which are very light and slow and already safe. One company, Starship, is the leader in that space — for disclosure I worked at Starship in its early years and hold shares – and it has now completed close to 4 million paid autonomous deliveries, and operates 2,000 robots, which is more trips and robots than any of other autonomous vehicle company, thanks to the simpler problem being solved.
Trucking companies are working at the other end, with big heavy vehicles, but limiting operations to simple rural interstates, which is again a more tractable problem. Aurora, which was one of the hottest startups working on a robotaxi, has pivoted now primarily to trucking, and half a dozen other ventures are hard at work on trucks. The main argument of those who feel robocars are far away is that there is too much complexity on the street, but that’s much less the case on the freeway. There, the problems are speed and stopping distance, and the extra level of safety needed due to the high weight of the vehicle. Gatik, on the other hand, is doing regular middle-mile operations for Loblaws and Walmart moving goods between warehouses and stores on a commercial basis.
Waabi, a year-old startup in Toronto taking a machine learning plus simulation approach just announced today that big-rig trucking will be their target.
Off-road automation
Even before Waymo began at Google, companies were making self-driving mining vehicles and farming tractors, and that effort continues, and is getting better and better. Robots will be roaming around warehouse lots, enclaves and campuses — and they already are. Several companies made shuttles and vehicles for places like retirement communities. This is not the grand market of car-replacement, but it still has value.
Freeway self-driving
When Google’s car project began, its first target was freeway driving, though it was working on every type of road. It was clear that while obviously scary because of the speed, the highway environment is much more controlled and simpler. It has no intersections, no pedestrians or bicycles, no oncoming traffic and no stop signs or traffic lights. It’s even easier than the trucking problem as you don’t need to deal with the issues of a giant, heavy vehicle.
The problem, of course, is that you have to get to and from the freeway. This means freeway driving can only be an add-on feature for regular cars. Indeed, this was the origin of the badly named “level 3” concept — a car which drives itself in a constrained environment like the freeway, but can summon a human driver to take over (with warning, not in emergencies) for areas that can’t be handled, like getting to and from the freeway.
This technology can be highly valuable to people who take long freeway drives, including commuters and people on road trips or going to vacation homes. It can give them back valuable time in their day. However, to be uncharitable that’s mostly “give a nice boon in luxury cars.” For those in robocars to change the world, that’s nice, but not as exciting as reshaping the future of transportation. It can provide the safety benefit (but ADAS can also provide a lot of that.)
The good news is that in time, such a feature moves down from luxury cars into ordinary cars.
The early version of this, already sold in limited ways, is a “traffic jam assist.” That’s a system that can drive in slow, congested conditions without human supervision. It’s mostly “follow the leader” at those slow speeds and doesn’t need long-range sensors and many other abilities. As soon as the traffic clears up and the car sees the leader is leaving, it can summon the human to take over, but it’s not an urgent situation — it just slows the end of the traffic jam if they are slow to respond. These is even more of a perk for luxury cars, and worse, can even make traffic worse by making traffic jams more tolerable.
A Freeway car is a full self-driving car, just limited to certain roads. However, because freeways don’t allow you to pull over and stop to switch from self-driving to manual mode, the idea of the “level 3” standby driver was proposed, since you need to have somebody take over while on the freeway before the manual-zone (off-ramps and construction zones) appears. Otherwise there is no need for the live transition and no confusion about mythical “levels.” If the vehicle can handle the entire freeway, you start to even be able to consider things like sleeper cars, even before you can have a robotaxi.
Remote driving
Several companies are trying to make remote driving work. That needs extremely good network connectivity and skilled drivers, but it’s not out of the question. It doesn’t save the cost of a driver while driving, but it does save it when not driving — and most driven services end up with the driver being idle a lot of the time. It also enables easy one-way movement of cars when the driver is virtual. This is not, at its base, an autonomous car technology, but it turns out that you want to merge the two, and use self-driving car tech to make the remote driving safer and robust against network failures. It’s possible to have self-driving that can safely handle 5 second outages but not the whole driving task, and it can also make sure a remote driver doesn’t do anything stupid because of outages or their more limited perception.
Whistlecar
Long ago I suggested the name “whistlecar” for a car that can be summoned to you in an autonomous mode, but you drive yourself once it gets to you. Some companies are doing this not in autonomous mode but using remote driving, such as Halo. This doesn’t give people back their time, but does allow carshare to be much more practical and creates new transport options, making it very interesting. It can be done today not just because of remote driving, but because the summoned car, like a delivery robot, doesn’t need to take every road or go very fast, and it can’t hurt passengers, so it can be built sooner.
There have even been efforts at self-delivering scooters, which solve micromobility’s problems — scooters left strewn on sidewalks, need for humans to recharge them and need to go find one — but both the efforts (I was involved in one) failed in the pandemic.
Robocar Airport
An airplane with 3 stairs and rows of robot pods at each one could be a different way to do an … [+]
One interesting option at the intersection of ground and air transportation is a robocar airport, that has few buildings and just takes passengers to and from and between aircraft, using all their doors. I outlined the Robocar airport in 2011 but it remains hypothetical — however the technology to make it is certainly ready today for a controlled environment like an airport.
Delivering the benefits
When we look at the core benefits of Robocars:
- Superior safety
- Give people back their time and eliminate the cost of a driver
- Allow cars to be delivered on demand for easy sharing, reworking how transportation happens
- Lowering the cost of short-hop rides (where human drivers are impractical, Uber minimum fare is $7 for one-mile ride) to enable new types of shared transport
- Allow cars to flexibly park themselves, valet-style in all forms of new spaces to eliminate the need for most parking lots.
- Allow cars to refuel or recharge themselves, removing all issues around electrification.
- Low cost delivery
We actually see that it is primarily #2 which requires the full robocar, though all of #3 to #6 only become really cheap when you have it. Even so, they are possible at a better cost with cars with remote driving or whistlecars. Safety remains #1, but if ADAS can provide it, a lot of these benefits can arise even before the full robotaxi is ready.
Cost is pretty important though. To take over the world, robotaxis have to compete with and beat car ownership to get people to switch. That’s a bit harder if human drivers have to be paid to deliver them, park them, and charge them. If people still have to drive them, the great benefit of an extra hour of time in your day is lost.
But either way, lots of interesting changes are coming to transportation, even if you are a skeptic about real robocars zooming around unmanned and with passengers any time soon.
