Mercedes Drive Pilot Lets You Play Games In Traffic Jams, Is It A Big Deal?

Mercedes has released a new version of their “drive pilot” system in high-end cars that will let the driver ignore the road in certain special circumstances, namely some highway traffic jams. It will operate now in Germany, and they are hoping for regulatory approval in California and Nevada by the end of 2022. They call it a “Level 3” system, based on a misguided set of self-driving levels from NHTSA and SAE, and is one of the earliest such systems to be deployed by a car manufacturer. It’s pricey and fairly limited, though it’s notable that Tesla

TSLA
doesn’t do it. They demonstrated it recently as a possible $5K add-on to a S-Class 580. To understand it, there’s a lot to unpack:

1. The limitations of this system and when they might go away
2. The sensors and technology
3. What other players are doing and not doing
4. The meaning of the highly-flawed “levels” of self-driving
5. The value to customers and society

What it does

This is a car that can mostly drive itself, but only on limited access highways in good weather conditions when traffic is below 40mph — in other words, during a traffic jam. That means that if you are on the highway and traffic slows down, you can push a button, and take your hands off the wheel — and more importantly take your eyes off the road, allowing you to do things like watch videos and play games on the car’s console, or chat face to face with your companion. Oddly, Mercedes says that you can’t do what everybody wants to do, which is use your phone, because the laws currently forbid that. Of course, people will do that anyway — and much more they aren’t supposed to do, like fall asleep.

When traffic clears up and the car in front of you pulls far enough away, the car signals the driver to take over. They use sound and lights, and in a nice trick, tightening the seat belt to make sure you pay attention. You need to take over if your exit is approaching, or there is construction, or the weather changes as well. They also make a big deal that you must take over if an emergency vehicle is present, and they have sensors looking for flashing lights in all directions. It doesn’t work at night or in snow, there is conflicting information on how well it handles rain. No tunnels or toll booths or other such complex things.

Sensors on the vehicle include those cameras, plus radars, a forward LIDAR and some microphones to listen for rain and presumably sirens. LIDAR is just now showing up in consumer cars for driver assist. Famously, Tesla deprecates LIDAR, and not just because of its higher cost, though that’s the reason their hardware base didn’t include it. Almost all self-driving teams plan to use LIDAR for reliable sensing.

Mercedes also has put in some extra accurate GPS. That’s surprising because pretty much all self-driving teams use GPS sparingly. They consider it a clue to help figuring out location, but since it’s only a clue they don’t need to overdo it. It doesn’t work in urban canyons or tunnels as well. Mercedes Drive Pilot also doesn’t work in those locations which leads to speculation it might rely more heavily on GPS than other systems. It is true that modern differential high-accuracy GPS systems on open roads can be more accurate than GPS of the past.

The driver has 10 seconds to take over after a warning. While I have not seen details on this, most traffic-jam tools have taken advantage of the fact that it is fairly safe to just come to a stop in the middle of a lane during a traffic jam. The smarts to pull over are not needed as they might be if you want to drive at full speed.

What about Tesla and others?

A few other companies have worked on such “level 3” systems like Audi (which backed down) and Honda, which sold only a small number of Legends in Japan with this function. Of course, famously, Tesla has been selling customers a “Full Self Driving” package which is a promise that in the future, they will get that ability. A very simple and poorly performing prototype is now available to a large subset of those customers in what is called a “beta test” though the software is not yet remotely near beta level. Tesla’s older “Autopilot” system works only on highway situations (including at full speed) and works much better, but is still only a driver-assist system needing full time supervision.

Many have wondered why Tesla doesn’t, as an intermediate stage, release a traffic-jam pilot that does actual self-driving (no supervision needed) system, similar to this Mercedes. While there are merits of going for the brass ring without distraction, there is also merit in giving a useful feature to customers who paid a lot of money. (In describing the price of their system, Mercedes has only said it will cost well under $12,000 — the current price of Tesla’s promise of a future system.)

Mercedes promotes their use of LIDAR. Tesla believes LIDAR is a distraction, and that the only solution is to develop advanced AI that doesn’t need LIDARs. Other teams like LIDAR for its high reliability — it’s not going to ever miss (or falsely identify) something as big as a car 200 feet away from you in a traffic jam. The reliability is “bet your life” level, unlike that of computer vision. For low-speeds and short ranges though, computer vision can do the job — especially stereo vision, though Tesla doesn’t have that — and so they should be able to play.

“Level Three” and liability

A key thing about real self driving (with standby driver or otherwise) is that it does mean the self-driving system takes responsibility for mistakes, while with ADAS there is a supervising driver with responsibility. This is needed because you are not going to hand off driving to a system if you will be blamed if it makes a mistake.

In 2011, the US safety agency NHTSA laid out a document with “levels” of self-driving. They weren’t developers and really didn’t understand the problem so their levels were very poorly laid out, but they caught on and were adapted by the Society of Automotive Engineers (who also are not leading self-driving developers.) While some of us protested that the levels were inaccurate and not really levels — because they did not describe a likely way in which the technology would progress — they came into wide use. The levels categorized self-driving systems by the role of the human being involved, a bit like categorizing motorized vehicles by the role of the horse.

The most debated of these levels was the 3rd, where a car could drive itself, but only on the highway, and so a human driver had to be on “standby” to take over when the highway ended or there was construction or something else the system could not handle. This was different from a monitoring driver who has to take over in urgent situations (such as the car trying to drive off the road.)

Some viewed this as a technical “level” — the car isn’t good enough to handle all the roads so it needs the standby driver. Others viewed it as a legal matter — a level 3 car was just a level 2 car so good that the maker could take legal responsibility for what it did. Those who felt there was no progression from ADAS to self-driving didn’t feel it made sense to talk about level 2 getting that good, but it is true that a real self-driving car does need to take legal responsibility. You would not get into an Uber

UBER
if told you would have to pay or even go to jail if your Uber driver was negligent.

So while some see this level 3 as a progress from level 2, it’s actually a special subset of a self-driving car (level 4.) The very particular subset is a car that needs to leave its driving domain while moving, which essentially is primarily on the highway. Every other type of self-driving car (except the mythical level 5 which does not actually exist) has a domain in which it works. It doesn’t leave that domain. It may come to a stop at the edge of the domain and then turn into a manually driven car. The level 3 car is special, in that the domain can end at a freeway off-ramp or the start of a construction zone, and there it must plan for a handoff while it is moving. But otherwise, it is a full self-driving car within its particular domain (freeways) and must handle any contingency in that zone.

The handoff

The handoff idea was always challenging. Most people settled on needing about 10 seconds of warning. Studies showed that almost always people could stop what they were doing and take over in 10 seconds. But not always — particularly if they have fallen asleep. They will fall asleep, unless you have major countermeasures against it.

Research done at Google

GOOG
with employees commuting in such cars in 2013 found that even paid employees could not be trusted to follow the rules, so they gave up any effort towards the idea of a standby driver. Others took the plan of just coming to a stop in the road, or being able to pull off if the driver didn’t respond within the 10 seconds. Stopping in the middle of the road is somewhat risky, but usually works — and is pretty much always safe in a traffic jam.

The value

Google decided not to make such a product both because of the challenge of making it safe and the question about its value. For many drivers, to be able to relax on the highway would certainly be valuable. Doing it just in traffic jams is not nearly as valuable, but still a very pleasant feature. Both of these functions, however, only allow for making private cars — starting with private luxury cars — a bit more useful and valuable. While they should improve driving safety over time, the initial ones will be similar to humans in safety, and the rate of crashes on freeways is actually much lower than on city streets.

The real value comes in the robotaxi concept for many. The robotaxi changes the nature of transportation and the city, eliminating the need for parking and for car ownership. Robotaxis can refuel/recharge themselves as well. That’s what revolutionizes industries. The big teams at tech companies want to do something big, not just improve luxury cars. If this feature does cost $5,000 as expected, it should make sales in that space, though that’s a fairly high surcharge for ordinary cars. Even so, once it can do the whole commute, people with long commutes or vacation homes might easily pay that price.

One downside of traffic-jam assists is that they make traffic jams more tolerable, which could actually increase them. Today, many drivers avoid driving in rush hour because they hate traffic jams. If a jam becomes a chance to get some work done or even to sleep, more people might be ready to drive when traffic is thick. They might even welcome the traffic jam, particularly in electric cars which are efficient in that situation.

The Mercedes offering is somewhat strange in how they tell you to only use applications on their console. Like most car OEM infotainment systems, these are far inferior to what you get on your phone or tablet. I don’t think most customers will obey that restriction, unless they are forced by a camera that watches them.

Naturally, Mercedes will go further. German automakers have tended to be conservative on deploying technology like this, and even pulled back on their actual robocar projects, but we should expect this one to grow, though German regulations will need to grow with it. Before long we might see the ability to go at full highway speed.

There, Germans have an edge because many of their freeways have no official speed limit. In other countries, the system designers will be forced to deal with the issue of deliberately going over the speed limit if they are to move at the speed of other traffic and not be an obstruction. On the plus side, systems like this can have the speed set by the human driver, who will be responsible for the speeding ticket, but it could be the developers will be afraid of letting the driver set it at the speed of traffic. We just saw NHTSA force Tesla to remove the ability to set their system to do rolling stops at completely deserted 4-way stops, something that is 100% safe, and done constantly by human drivers, though usually illegal in the USA. (In Europe, 4 way stops are quite rare, and they use roundabouts and yields instead, with better safety records!) For the first time NHTSA was forcing cars to obey the law, rather than just be safe.

In the USA, there is no legal requirement for the 40mph limit, but then there is no legal definition of how to get approval to sell the full-speed car either. In full speed traffic, just stopping in the middle of the lane may not cut it, either. But if the cars have to stick to the speed limit, they will both be a little dangerous and create a lot of frustrations on the road. We’ll find out the answer soon.