Robocars enter phase two as different strategies abound

Topic: 

At CES 2018, autos took over the show, and self-driving took over autos. At least in the industry, it's now mainstream. So what new approaches are teams taking, and how do they hope to win?

Waymo remains the leader by a very large margin, though not all agree with that. They were not present at CES, though Google was for the first time in many years. Most car companies had mention of self-driving plans but not too much to show specifically. That's because car companies are not ready to sell anything to customers yet. In fact, BMW's booth was all about driving and racing, with people lining up to drift in the parking lot. Suppliers, on the other hand, want to sell to car companies, so in spite of the word consumer in the show, they were there.

It is no longer so impressive to have a decent self-driving stack working at some level. Many suppliers, car companies, tech companies sand startups have this. It's easy to find demos of a car driving around some urban streets now. Not too long ago, simply being able to do that was impressive.

Now the key difference, which is much harder to see, is the difference between being able to drive for a day, and the ability to drive without incident for 20 years.

(The average human drives around 100,000 miles in 10 years, and according to recent estimate probably has one minor "ding" accident every 100,000 miles. An accident reported to police happens every 500,000 miles or 50 years. We all hope that robocars will be able to beat these human numbers by a decent margin when they are released.)

So while the most press was around Aptiv (Delphi) and Lyft's demo where you could summon a Lyft in the gold parking lot and take it to a special set of destinations, that demo simply says, "we are in the game, too." (Though to give credit, I think Aptiv is one of the more advanced suppliers when it comes to their thinking.)

This is why I identified Waymo's removal of safety drivers from behind the wheel as the top robocar story of 2017. It's the most public declaration of reaching such a high new safety level, even if only in Phoenix, which is one of the easiest cities in the world to drive. Other companies have announced that they hope to do this in 2018 -- when they do it will put them ahead of most of the pack.

Auto Industry

There are too many teams to track, but they are falling into these broad classes:

Car company planning to sell luxury robocar

At first this was almost every car company plan -- to make a vehicle with self-driving as an add-on feature. Due to the cost, this is almost surely an expensive feature on an expensive car. Customers would own these cars, and they would be sold in the standard way.

Most car companies still plan an offering of this sort, though Ford has declared its cars are only for taxi fleet operations. Most offerings begin with autopilots, and then traffic jams with a standby driver, then highway with a standby driver and finally full unmanned-capable operations, at least at low speed. Owned cars don't need nearly as much ability to drive long trips on their own, and in fact the plan for most is that the human drives a moderate amount, but has a button to press to get into self-driving mode.

Most of the thinking at these companies can be classified as "evolve our ADAS work until we get self-driving" which I have argued is very much the strategy that will doom some companies.

Car company planning to make a taxi fleet car

Ford was the first to declare this intention, though Mercedes and BMW both talked about it in the context of their one-way car sharing services DriveNow and Car2Go. Now others have joined in. A taxi fleet car can actually be more expensive than a car-sales car, since it works such a long duty cycle. Add $10,000 to the cost of a car and it's very hard to sell that in the dealership, especially with the large markup needed over the Bill-of-Materials (BOM) cost. Add it to a taxi that goes 250,000 miles and it's only adding 4 cents/mile to the underlying cost in a taxi ride.

Note that most car companies planning a luxury car also have thoughts of taxi fleet operations, either run by them or their customers. Tesla has spoken of how it plans a service where people can hire out their autonomous Tesla when not using it. An interesting question is whether the companies that take a "retail car first" strategy will be able to catch up in the taxi market with those taking a taxi first strategy.

Car companies that really have little idea what they are doing

There are companies that have announced almost nothing, or indeed seem to be doing nothing. There are car companies whose announcements have all been either silly gimmicks or "partnerships" with nothing real underneath them. I am not going to name there here to be nice.

Top tier auto suppliers

The suppliers fall into a couple of different groups. They all want to be the supplier of the self-drive system, or at least components for it, including sensors and "ECUs" (dedicated embedded controls for specific functions.) Auto suppliers and car makers tend not to think of themselves as software companies, and they definitely don't think like software companies. Everybody wants to sell hardware units that include processing in the hardware, and expect a car to be built up from several of these. This is a bit at odds with the more whole-system design seen outside car companies, where the master software architecture is the big project, though it will rely on self-contained hardware units, in particular sensors.

The typical radar sold by an auto supplier has processing inside it which returns a list of the radar targets it is seeing at any given time. For a long time, it was very difficult to get the raw radar data from them if you wanted to do your own processing. The MobilEye's normal operation was as a processor that took that camera data and provided high level outputs.

It is quite normal for advanced features in cars to be made by the auto suppliers, and most ADAS tools came from this market. At the same time, companies now realize that the self-driving system is a major differentiator, where adaptive cruise control and collision warning were not. This means they feel much more strongly that they should own it, and not buy the core value from a supplier. As such, most car companies have their own large teams now, which means the suppliers will sell only to the "me too" companies that failed at making their own team.

Suppliers will also sell components, including sensors, software and "hardware modules" that are really software packaged as hardware -- as is common in the auto supplier world. Even the automakers demanding to own the central system will be happy to buy components.

High-tech players

The large company high-tech players, such as Google/Waymo, Apple, Baidu, Samsung and Naver are well known. There are also the chip companies, most notably Nvidia and Intel/MobilEye. The push by chip companies has been interesting. They are always looking for new markets for their chips, and it's common for them to pre-build reference versions of applications to kickstart that. Cars will never be sold in the numbers that mobile phones or even laptops are sold. The chips have to be much more expensive, and they can be, but there are limits.

While the chip companies will sell chips and this justifies their efforts, I am a bit wary of those who have been boosting the stock prices of these companies greatly due to these efforts. I believe that robocars will be a highly dynamic and constantly changing field, which means a software dominated field. New hardware generations are expensive and take time. new software

Startup players

The field is now rife with startups, including "full stack" startups that are putting together all the tools (except the car and major hardware) to make a taxi, and software component startups which are making specific modules for perception or mapping or other elements they hope all the other players will need. There's a bit less of this in the software space -- indeed at CES I talked to Dibotics, a small company which makes software to help process LIDAR point clouds, and they were surprised they did not have more competitors. It could be that the full stack players simply feel they want to own this part of the system.

At CES I saw AIMotive again, the company form Hungary. They have a large team but have decided to be one of the few pushing at building a driving stack that uses cameras and lot LIDAR. They join Tesla, comma.ai and a few other minor players taking that gamble.

Earlier I covered the many hardware startups, particularly in LIDAR and other sensors. This space makes a lot of sense, and a lot of people figured that out from the start.

Delivery players

I"m involved with Starship, the first of the players in the delivery robot space. Recent times have shown a lot of competitors show up in this space, a few of which were at CES. Most are small at present, but some recent entrants such as Nuro are well funded and plan to operate on streets rather than sidewalks.

Local Motors

An interesting play that deserves there own category is Local Motors. I'll write about them in a future post.

Comments

If platooning robotaxis had a physical coupling that came out of the bumper like a train, then the cars on short range journeys could power the cars on long range journeys giving them an indefinite range.

Such physical coupling is actually quite difficult and also requires standardization of connectors handling high loads and in the case you describe, very high currents. Not an easy thing to do. And the value is limited. Robotaxi journeys (certainly long ones) will all have their destination known in advance, and so if you say you are going 300 miles, you will get a car ready to go 300 miles. You would not be given a short range car. Probably you will get a liquid fuel car which has the option of fast refueling, if a long range electric vehicle is not available. Or you can have the option not of recharging, but simply swapping to a new vehicle with more range -- it will take only 30 seconds to swap.

With all that possible, why try to build a complex and dangerous standard in a fast changing world?

Recently, a claim was made in IEEE Spectrum, https://spectrum.ieee.org/cars-that-think/transportation/self-driving/the-selfdriving-cars-bicycle-problem that self-driving cars have a bicycle problem. And, presumably, a motorcycle problem.

In your opinion, is this legit?

--dave
Dave Brown, back at Waterloo, Collier-Brown after I married.

I don't doubt that bicycles present problems. This article talks almost entirely about the problems for computer vision systems. This is one reason only a tiny minority of teams wish to rely on computer vision. Most use lidar, which sees bicycles similarly to how it sees cars. Ditto radar but of course that has very low resolution -- but it turns out the spinning legs and wheels of a bike have a special radar signature it is easy to spot.

It is nonetheless surely true that since bicyclists are extra vulnerable that extra attention is needed on them. The article suggests they can move unpredictably, and they can, but there are two classes of such movement. If they move in a legal manner, along their right of way, they will not be at much risk from the robocars which will keep to their own right of way fairly religiously. There is more concern if the bicycle suddenly incurs on a car's ROW. Of course that is legally the cyclist's fault but systems will still want to try and be as safe as they can even in that situation. Some cars already are programmed to give as wide a berth as they can legally do to reduce the risk of surprise bike actions.

There is also the issue of the bike which appears "out of nowhere" from behind things like a sign, van or other obstacle. One program, Citimobil2, goes absolutely paranoid about this and keeps vehicles to super low speeds so this can never happen. That is not practical for serious vehicles, however.

As a cyclist who was left for dead last year after entering a car's ROW (which is precisely the locations where cars can leave you for dead; the law didn't matter in my case and never does) I have a different perspective. Bicycles are actually the easiest problem to solve. There is a wonderful demonstration project of the technology that completely cures this problem; it is called the Netherlands. Yup, you gotta keep em separated! Problem solved.

They intersect all the time in the Netherlands, but only at intersections of course. One nice thing about robocars, which is that you can make a virtual bike lane. The robocar will not enter it and the bike can be safe in it. You don't have to build the physical lane like they do in the Netherlands.

I followed the link to the local motors website, sadly I came away none the wiser about what their vision was and where there device fitted into the brave new driverless world.
I'm assuming it's designed as a last mile connector to mass transport, or for short trips to the shops?
If so I hope it succeeds, though that may of course depend on the technology rather than the concept.

Anyway, I'm looking forward to your piece on it to get a better understanding of what it is.

Add new comment