My recent article on a future vision for public transit drew some ire by those who viewed it as anti-transit. Instead, the article broke with transit orthodoxy by suggesting that smaller vehicles (including cars and single person pods) might produce more efficient transit than big vehicles. Transitophiles love big vehicles for reasons beyond their potential efficiency, so it's a hard sell.

Let's look at the factors which determine what vehicle size makes the best transit.

Perhaps the world's most exciting new technology today are deep neural networks, in particular the convolutional neural networks such as "Deep Learning." These networks are conquering some of the most well known problems in artificial intelligence and pattern matching, and since their development just a few years ago, milestones in AI have been falling as computer systems that match or surpass human capability have been demonstrated. Playing Go is just the most recent famous example.

This is particularly true in image recognition. Over the past several years, neural network systems have gotten better than humans at problems like recognizing street signs in camera images and even beating radiologists at identifying cancers in medical scans.

These networks are having their effect on robocar development. They are allowing significant progress in the use of vision systems for robotics and driving, making those progress much faster than expected. 2 years ago, I declared that the time when vision systems would be good enough to build a safe robocar without lidar was still fairly far away. That day has not yet arrived, but it is definitely closer, and it's much harder to say it won't be soon. At the same time, LIDAR and other sensors are improving and dropping in price. Quanergy (to whom I am an advisor) plans to ship $250 8-line LIDARS this year, and $100 high resolution LIDARS in the next couple of years.

The deep neural networks are a primary tool of MobilEye, the Jerusalem company which makes camera systems and machine-vision ASICs for the ADAS (Advanced Driver Assistance Systems) market. This is the chip used in Tesla's autopilot, and Tesla claims it has done a great deal of its own custom development, while MobilEye claims the important magic sauce is still mostly them. NVIDIA has made a big push into the robocar market by promoting their high end GPUs as the supercomputing tool cars will need to run these networks well. The two companies disagree, of course, on whether GPUs or ASCICs are the best tool for this -- more on that later.

In comes comma.ai

In February, I rode in an experimental car that took this idea to the extreme. The small startup comma.ai, lead by iPhone hacker George Hotz, got some press by building an autopilot similar in capability to many others from car companies in a short amount of time. In January, I wrote an introduction to their approach including how they used quick hacking of the car's network bus to simplify having the computer control the car. They did it with CNNs, and almost entirely with CNNs. Their car feeds the images from a camera into the network, and out from the network come commands to adjust the steering and speed to keep a car in its lane. As such, there is very little traditional code in the system, just the neural network and a bit of control logic.

Here's a video of the car taking us for a drive:

The network is built instead by training it. They drive the car around, and the car learns from the humans driving it what to do when it sees things in the field of view. To help in this training, they also give the car a LIDAR which provides an accurate 3D scan of the environment to more absolutely detect the presence of cars and other users of the road. By letting the network know during training that "there is really something there at these coordinates," the network can learn how to tell the same thing from just the camera images. When it is time to drive, the network does not get the LIDAR data, however it does produce outputs of where it thinks the other cars are, allowing developers to test how well it is seeing things.

This approach is both interesting and frightening. This allows the development of a credible autopilot, but at the same time, the developers have minimal information about how it works, and never can truly understand why it is making the decisions it does. If it makes an error, they will generally not know why it made the error, though they can give it more training data until it no longer makes the error. (They can also replay all other scenarios for which they have recorded data to make sure no new errors are made with the new training data.)

I frequently see people claim that one effect of robocars is that because we'll share the cars (when they work as taxis) and most cars stay idle 95% of the time, that a lot fewer cars will be made -- which is good news for everybody but the car industry. I did some analysis of why that's not necessarily true and recent analysis shows the problem to be even more complex than I first laid out.

To summarize, in a world of robotic taxis, just like today's taxis, they don't wear out by the year any more, they wear out by the mile (or km.) Taxis in New York last about 5 years and about 250,000 miles, for example. Once cars wear out by the mile, the number of cars you need to build per year is equal to:

Total Vehicle Miles per year
Avg Car Lifetime in Miles

As you can see, the simple equation does not involve how many people share the vehicle at all! As long as the car is used enough that the car isn't junked before it wears out from miles, nothing changes. It's never that simple, however, and some new factors come into play. The actual model is very complex with a lot of parameters -- we don't know enough to make a good prediction.

People travel more in cars.

It's likely that the number of miles people want to travel goes up for a variety of reasons. Robocars make car travel much more pleasant and convenient. Some people might decide to live further from work now that they can work, read, socialize or even sleep on the commute. They might make all sorts of trips more often. Outside of rush hour, they might also be more likely to switch from other modes, such as public transit, and even flying. Consider two places about a 5 hour drive apart -- today flying is going to take just under 3 hours due to all the hassles we've added to flying, even with the improvements robocars make to those hassles. Many might prefer an uninterrupted car ride where they can work, watch videos or sleep.

Vehicles run empty to reposition

Regular taxis have wasted miles between rides. Indeed, a New York taxi has no passenger 38% of the time. Fortunately, robocars will be a lot more efficient than that, since they don't need to cruise around looking for rides. Research suggests a more modest 10% "empty mile" cost, but this will vary from situation to situation. If you need the robotaxi fleet to constantly run empty in the reverse commute direction, it could get worse. Among those who believe robocars will be more personally owned than used as taxis, we often see a story painted of how a household has a car that takes one person to work, and returns home empty to take the 2nd person, and then returns again to take others on daytime errands. This is possible, but pretty inefficient. I think it's far more likely that in the long term, such families will just use other taxi services rather than have their car return home to serve another family member.

Cars last longer

The bottom part of the equation is likely to increase, which reduces the number of cars made. Today, cars are engineered for their expected life-cycle -- 19 years and 190,000 miles in California, for example. Once you know your car is going to have a high duty cycle, you change how you engineer it. In particular, you combine engineering of parts for your new desired life cycle with specific replacement schedules for things that will wear out sooner. You want to avoid junking a car with lots of life in the engine just because the seats are worn out, so you make it easy to replace the seats, and you have the car bring itself to a service center where that's fast and easy.

General Motors has purchased "Cruise," a small self-driving startup in San Francisco. Rumours suggest the price was over one billion dollars. In addition, other rumours have come to me suggesting that at least one other startup has been seeking a new round of funding at that valuation, but did not succeed due to the market downturn.

Reports released reveal that one of Google's Gen-2 vehicles (the Lexus) has a fender-bender (with a bus) with some responsibility assigned to the system. This is the first crash of this type -- all other impacts have been reported as fairly clearly the fault of the other driver.

I have a big article forthcoming on the future of public transit. I believe that with the robocar (and van) it moves from being scheduled, route-based mass transit to on-demand, ad-hoc route medium and small vehicle transit. That's in part because of the disturbingly poor economics of current mass transit, especially in the USA. We can do much better.

I recently read a report of a plan for a new type of intersection being developed in Malaysia, and I felt it had some interesting applications for robocars.

The idea behind the intersection is that you have a traditional intersection, but dig in one or both directions, a special underpass which is both shallow and narrow. One would typically imagine this underpass as being 2 vehicles wide in the center of the road but other options are possible. The underpass might be very shallow, perhaps just 4 to 5 feet high.

I've been electric car shopping, but one thing has stood out as a big concern. Many electric cars are depreciating fast, and it may get even faster. I think part of this is due to the fact that electric cars are a bit more like electronics devices than they are cars. Electric cars will see major innovation in the next few years, as well as a decline in their price/performance of their batteries. This spells doom for their value. It's akin to cell phones -- your 2 year old cell phone still functions perfectly, but you dispose of it for a new one because of the pace of innovation.

While I've been in love for a long time with the idea of mobility-on-demand and the robocar taxi, I continue to have some privacy concerns. The first is simply over the idea that a service company gets a map of all your travels. Of course, your cell phone company, and companies like Google with their Location History (Warning, don't click or you will be freaked out if you didn't know about this) know this already, as does the NSA and probably all the other spy agencies in the world. That doesn't make it much better to add more trackers.

NHTSA, the federal car safety agency has been talking about getting into the robocar game for a while, and now declares it wants more involvement with two important details:

Hot on the heels of my CES Report is the release of the latest article from Chris Urmson on The View from the Front Seat of the Google Car. Chris heads engineering on the project (and until recently led the entire project.)

I'm back from CES 2016 with a raft of news, starting with robocars. Some news was reported before the show but almost everybody had something to say -- even if it was only to have something to say!

I have many more photos with coverage in my CES 2016 Photo Gallery.

Ford makes strong commitment

Ford's CEO talks like he gets it. Ford did not have too much to show -- they announced they will be moving to Velodyne's new lower cost 32-laser puck-sized LIDAR for their research, and boosting their research fleet to 30 vehicles. They plan for full-auto operation in limited regions fairly soon.

Ford is also making its own efforts into one-way car share (similar to Daimler Car2Go and BMW DriveNow) called GoDrive, which pushes Ford more firmly into the idea of selling rides rather than cars. The car companies are clearly believing this sooner than I expected, and the reason is very clearly the success of Uber. (As I have said, it's a mistake to think of Uber as competition for the taxi companies. Uber is competition for the car companies.)

Ford is also doing an interesting "car swap" product. While details are scant, it seems what the service will do is let you swap your Ford for somebody else's different Ford. For example, if somebody has an F-150 or Transit Van that they know they won't use the cargo features on some day or weekend, you drive over with your ordinary sedan and swap temporarily for their truck -- presumably with a small amount of money flowing to the more popular vehicle. Useful idea.

The big announcement that didn't happen was the much-rumoured alliance between Ford and Google. Ford did not overtly refute it but suggested they had enough partners at present. The alliance would be a good idea, but either the rumours were wrong, or they are waiting for another event (such as the upcoming Detroit Auto Show) to talk about it.

Faraday Future, where art thou?

The big disappointment of the event was the silly concept racecar shown by Faraday Future. Oh, sure, it's a cool electric racecar, but it has absolutely nothing to do with everything we've heard about this company, namely that they are building a consumer electric car-on-demand service with autonomous delivery. Everybody wondered if they had booked the space and did not have their real demo ready on time. It stays secret for a while, it seems. Recent hires, such as Jan Becker, the former head of the autonomous lab for Bosch, suggest they are definitely going autonomous.

Mapping heats up

Google's car drives by having super-detailed maps of all the roads, and that's the correct approach. Google is unlikely to hand out its maps, so both Here/Navteq (now owned by a consortium of auto companies in Germany) and TomTom have efforts to produce similar maps to licence to non-Google robocar teams. They are taking fairly different approaches, which will be the subject of a future article.

One interesting edge is that these companies plan to partner with big automakers and not just give them map data but expect data in return. That means that each company will have a giant fleet of cars constantly scanning the road, and immediately reporting any differences between the map and the territory. With proper scale, they should get reports on changes to the road literally within minutes of them happening. The first car to encounter a change will still need to be able to handle it, possibly by pulling over and/or asking the human passenger to help, but this will be a very rare event.

MobilEye has announced a similar plan, and they are already the camera in a large fraction of advanced cars on the road today. MobilEye has a primary focus on vision, rather than Lidar, but will have lots of sources of data. Tesla has also been uploading data from their cars, though it does not (as far as I know) make as extensive use of detailed maps, though it does rely on general maps.

Lyft announced a $500M investment from GM with $500M more, pushing them to a $5.4B valuation, which is both huge and just a tenth of Uber. This was combined with talk of a push to robocars. (GM will provide a car rental service to Lyft drivers to start, but the speculation is that whatever robocar GM gets involved in will show up at Lyft.)

Yahoo Autos is reporting rumours that Google and Ford will announce a partnership at CES. Google has always said it doesn't want to build the cars, and Ford makes sense as a partner -- big, but with only modest R&D efforts of its own, and frankly a brand that needs a jolt of excitement. That means it will be willing to work with Google as a partner which calls many of the shots, rather than just viewing them as a supplier, which gets to call few of them.

Be careful what you wish for -- yesterday the California DMV released its proposed regulations for the operation of robocars in California. All of this sprang from Google's request to states that they start writing such regulations to ensure that their cars were legal, and California's DMV took much longer than expected to release these regulations, which Google found quite upsetting.

Another road trip has meant fewer posts -- this trip included being in Paris on the night of Nov 13 but fortunately taking a train out a couple of hours before the shooting began, and I am now in South Africa on the way to Budapest -- but a few recent items merit some comment.

I'm pleased to announce today the unveiling of a new self-driving vehicle company with which I am involved, not building self-driving cars, but instead small delivery robots which are going to change the face of retailing and last-mile delivery and logistics.

In the buzz over the Tesla autopilot update, a lot of commentary has appeared comparing this Autopilot with Google's car effort and other efforts and what I would call a "real" robocar -- one that can operate unmanned or with a passenger paying no attention to the road. We've seen claims that "Tesla has beaten Google to the punch" and other similar errors. While the Tesla release is a worthwhile step forward, the two should not be confused as all that similar.

Last night, one day early, I attended Stanford's unveiling of their newest research vehicle for self-driving. In order to do experiments with drifting (where you let the rear wheels skid freely) they heavily modified an old Delorean.

They managed to get Jamie Hyneman of Mythbusters to host the event so there was a good crowd. He asked "Why a Delorean?" and instead of saying the obvious line:

"The way I see it, if you're going to build a self-driving drifting car, why not do it with some style?"