Mercedez-Benz has announced approval of their “Drive Pilot” system, in Germany, which does fully autonomous operation in highway traffic jam situations. Such systems have been promised for many years. It’s interesting to note that Tesla TSLA +1.3%, which might be able to produce such a system today, has instead put all their focus on their poorly named “full” self-driving product, possibly at the expense of other features like this.
The German certification is designed to be extended to the rest of the world, starting of course with Europe. Honda has sold a modest number of Legend sedans in Japan with this function, and Audi produced one but did not finish the regulatory approval process.
A system like this sometimes called a “level 3” system though those levels are misleading and not typically used by leading self-driving developers. It got that number as a potential half step between driver-assist (Such as Tesla FSD and other Mercedes offerings) and real self-driving. It was dreamed up by regulators with no experience in developing self-driving systems and there’s been a lot of disparate thinking about it over the years.
The early conception was a full self-driving car that, because it wasn’t quite good enough to handle some situations, might call upon a driver who was on standby but not watching the road. That driver would have around 10 seconds to take over when something like a construction zone was approaching or the vehicle was coming to the end of the territory it could drive in. It was usually imagined the car would handle the highways at full speed, but need the standby driver to handle construction, take exits and get to and from the highway.
There are problems with this approach, which led to the idea of the “traffic jam” self-driving car. That’s a car that can do the full driving task only in a traffic jam situation, letting the driver relax and use their phone or computer. Driving in a traffic jam is easy — the speeds are slow, and you are surrounded by other cars and can follow what they do. If the jam ends and it’s time to speed up, the car can wait for the standby driver to take over to handle that.
Google GOOG +0.4% tried to build the highway standby system back in 2013 and tested it with employees, ordering them to keep attention to the road. It failed — the employees ignored the road, and on rare occasions even fell asleep. Google (now Waymo) decided this was not a worthwhile path.
From one direction, “level 3” can be treated as more a special version of level 4 which is real self-driving. Self driving always comes with a subset of roads and conditions in which it operates. Early level 3 is really a car which is level 4 in a special domain (highways) but by law is forced to travel at speed out of that zone (ie. off-ramps and construction.) If a car can come to a safe stop when near the edge of its operating zone and let a human driver take over from a stop, that’s a real self-driving robocar (level 4.) If the human has to take over while moving, that’s when you get this special case they thought was level 3.
There is an alternate definition though. This definition argues that level 3 is just the first system where the provider of the driving system (or carmaker) takes legal responsibility for driving. That it says to the former driver, “you are now a passenger” in this limited area.
Whether legal or technical or dangerous, nobody has made a serious stab at the first version of level 3, the highway car that needs occasional assist in known tough spots. The Mercedes car provides the traffic jam assist function — only on German motorways to start below 60 km/h. While people debate whether they want to drive their car or not, nobody likes driving in a traffic jam, and everybody hates the time wasted in them. As a luxury feature, this will let drivers make more productive use of that time. It’s a nice-to-have, not world changing, but a useful luxury.
It has a potential downside. By making traffic jams more palatable, it could encourage people to drive at rush hour when they otherwise might work to avoid it. They might actually look forward to the traffic jam, offering a chance to get some work done. This could even make traffic jams worse. Fortunately, other functions will come in the years ahead to reverse that.
Why doesn’t Tesla have it?
Tesla has been putting most of their development effort of late into their promised “full self driving” product. As it stands, the product is ludicrously far from full self driving to merit that name — it is still a driver assist system — but a beta version has been released to a limited number of Tesla drivers who test it and post videos. It is impressive for being able to act with only limited maps, though it frequently makes mistakes maps would prevent. Tesla has promised that the FSD product would be available soon for many years, and has demonstrated it has no ability to accurately predict its progress. That’s not surprising — for it to work requires breakthroughs in computer vision which do not yet exist, and whose arrival is difficult to predict.
Tesla has also been doing far less in the last 2 years in the way of new features in their software updates. In the earlier years of Tesla, cars would receive remarkable new features on a fairly regular basis. One of those features was the first Autopilot itself, and they have also included surprising features for a software update such as the cars becoming faster or getting more range. More recently the trail of new features has slowed greatly, and it suggests the speculation that the large effort to make the FSD product meet its arbitrary deadlines has distracted from other improvements in the vehicle.
A traffic jam assist is an improvement which would have immediate value, and which would make use of all the FSD research. While Tesla cars lack the LIDAR the Mercedes above has, and now lack the radar too, they could probably provide a traffic jam pilot function below a usable speed just with cameras.
Another useful feature would be better fully automated parking. Tesla released a product called “smart summon” which was supposed to let you summon the car to come to you through a parking lot. It received widespread derision — parking lots are actually one of the harder problems, even though speeds are low. Even so there are things Tesla could have done of value, such as automated valet parking in designated lots (Mercedes built such a system in Stuttgart but it was disappointing.) Some work could make this more valuable, and if the cars did true valet style parking by densely packing the cars together they could actually use less parking space and allow this parking to be cheaper than regular parking.
A highly useful function would be automatic operations at Tesla Superchargers. Sometimes these chargers fill up, and it would be very useful if a car, when done charging (or authorized by the driver) could allow the waiting driver to unplug it, and then move itself to an ordinary parking space to free up the station. Likewise, arriving drivers, rather than wait in a queue, could park close by and move into the next free charging spot (waiting for the next human who arrives or leaves to plug them in, which they would do both because it speeds up flow at the station, and to pay it forward for those who plug or unplug them when needed.) A similar function could happen at hotels with Tesla destination charging, or other parking lots with multiple chargers.
Super useful would be an ability for cars to drive slowly and late at night to chargers within a reasonable distance of their base. Driving passengers at full speed on busy daytime roads is challenging. 20 km/h operation with nobody inside at 2am is another story, and it’s doable today in many areas. Today, the charging lots serving such cars would need a modern equivalent of a “pump jockey” to do the plugging, but in the future, cars can and should develop a new plug mechanism (on the bottom, front or back of vehicles) which would allow cars to plug themselves in. (Being a robot, the car can do most of the movement needed to plug in with proper plug design.) While hundreds of billions are being spent on new charging infrastructure, the reality is that what owners would truly love is car that just magically is always fully charged because it wanders off when needed and refills. That can happen with much less infrastructure at much lower cost if the cars can make short trips to plug in.
Indeed, the easiest form of that (which even non-Tesla cars could do) would be the ability to do automatic plug-in at stalls in large parking lots and offices. Cheap to install, a small number of stations could handle many cars, as the average car only needs about 80 minutes of charging a day, on average, even at Level 2 speeds. With DC Fast at 100kw, each car only needs 6 minutes on average, and one station could refill over 70 cars in a 9-5 office day if they could plug themselves in.
(Inductive charging could also do this, but it’s an expensive addition to the car and less efficient.)
These are just a few of the things Tesla could do with their technology to bring positive results for customers sooner than they are likely to produce actual real full self-driving. Perhaps a change of priorities is in order.