Get best of the week

[Forecast] Will all unmanned trucks come to an end or only Starsky Robotics?

image

In March, Stefan Seltz-Axmacher, CEO and founder of Starsky Robotics, announced the completion of his shipping company, which combined the development of unmanned driving software and remote monitoring methods for moving large trucks ( Starsky Robotics unmanned trucks came to an end ). Back in 2015, I was advising Seltz-Axmacher on their project, but we lost touch. He recently published a detailed essay on the reasons for the failure of the company, which caused some concern, because one of his claims is that the task is too complicated, and other companies will not cope. Some reject this claim and believe that Starsky failed due to the failures of the company itself and not the industry.

Let's look at some of the arguments given in the essay:

  1. Controlled machine learning does not meet the task of autonomous driving and can be very far from solving it.
  2. While in the world of cargo transportation there should be a rich harvest for robotic trucks, in fact, the industry is closed and not ready.
  3. Venture companies do not like to invest in a business that should include a traditional, less profitable and more capital-intensive component (for example, owning and managing a cargo delivery service).
  4. Everyone says that security is No. 1, but in reality it is not, and it does not attract attention or investment.

image : — -, 2007 , . ClariNet, Electronic Frontier Foundation Foresight Institute, Singularity University.

In fact, I will start with the aspects of paragraph 2 that I dug up during the discussion with Seltz-Axmacher last week. Starsky chose trucking because it is a poorly managed industry with great potential for improvement. There is an acute shortage of drivers. Drivers are also sometimes unreliable - as one client said, the robot is unlikely to get into a fight with people in the warehouse, or decide to stop in Las Vegas and spend several days in a strip club. This does not always happen in road transport, but it happens often enough so that the exclusion of such a factor is a big plus. Starsky planned to release their trucks only on the simplest routes, on short and medium distances along clean, open and free highways. If the problem arose due to traffic or weather conditions,the truck would just stop - after many hours, stops are normal for human drivers.

All of this also means that combining fully unmanned driving with occasional remote assistance can work on these easy roads. Their trucks would never change lanes on their own. They would not mind driving slowly behind other trucks. They would not mind anything. They felt that they had chosen the easiest driving task for themselves, except, of course, the problems of moving a 40-ton vehicle at high speed and the associated risks.

Current fleet operators are not technology pioneers. Indeed, it will be difficult to convince them to do something so radical, even taking into account the problem of finding good drivers, or even any drivers. Initially, Starsky planned to work remotely, leaving only on roads about which there is good data. In 2015, I told them that finding such roads would not be easy. Today it is more real, and the deployment of services such as Starlink DTLK will probably greatly simplify this task in the near future.

If existing operators are not pioneers of technology, then, according to Starsky, investors are afraid that in order to become a robotic freight company, they will have to become a freight carrier. Perhaps they are right - the investment habits of venture capital funds seem strange to the rest of the world. They expect to invest in a dozen interesting startups, each of which can become a shot, after which they will expect the remaining 11 to try like crazy, but they will burn out. Traditional business does not usually fit into this form.

So here we have an example in which Seltz-Axmacher is right - shipping is difficult to sell - although this did not prevent many other freight concepts from getting a lot of investment.

Is AI too complicated?


You will need to read the article to find out all the details, but it touches on one of the big questions of 2020 - how much more difficult is it to create AI for unmanned vehicles than people initially thought? In particular, there was a lot of enthusiasm (a certain part of it was just a hype) about the capabilities of deep neural networks.

Almost every automotive team has done a great job of creating training data sets for machine learning systems. This involves collecting data from the real world (images and clouds from lidars) and tagging them for AI training. This technique yielded amazing results, by the standards of standards that existed only a few years ago. The question is, is it good enough to provide a level of quality for unmanned driving, and when will it happen (and will it ever happen?)

Seltz-Axmacher correctly notes that it is quite easy to get impressive results in the early stages, and this leads people to the idea that full-fledged success is just around the corner. Several companies even tried to build unmanned systems with neural networks, which are black boxes in which you insert pixels from cameras and receive commands to control systems (steering wheel and pedals).

He is right that pure supervised machine learning is not enough now, and even in the future it will take a certain distance. Tesla's TSLA is betting that this is not the case, but most companies are trying to build hybrids that use machine learning along with other algorithms. They still believe that this strategy will be successful. Usually they look with disdain on those who hope to take a holistic approach for the same reasons as Seltz-Axmacher. In 2019 and 2020, some market players retreated, especially those who did not pay attention to this gap in the automotive industry. Parties that influence this issue expect the gap to be wide. I think their efforts are in vain.

All this applies to the world of unmanned trucks. Cargo transportation attracts many companies, because driving on the highway is simple, even considering that the trucks are fast and weigh a lot. When I talk about simplicity, I mean, compared to city streets. Commercial value is also very clear. For that matter, the commercial value is too obvious, and in case of accidents (albeit at a lower cost), the opposite reaction may occur, namely that people are injured only in order to make transportation more efficient, and not for in order to change the operation of the transport as a whole.

Venture funds do not invest in such a business


This statement is also true, although not completely. There are promising venture capital companies and strategic investors who could invest in a more capital-intensive, infrastructure and intensive business. True, given the choice, they would prefer to invest in Uber, which writes only software and does not own cars. Profitability is much higher. But even Uber can get an investment by selling the story of the transition to ownership of large fleets of robotaxi with the replacement of drivers with AI.

Perhaps the fact is that with the onset of market slowdown and tremor, Starsky, and not their concept, could not stand the plan. Of course, several other companies increased their momentum and got good grades, although perhaps not as huge as a couple of years ago. The big prize is still being raffled off.


Seltz-Axmacher touches on a real problem - he wonders how much people really care about security. In the end, each company in each of its presentations says something like the following: “We all care about security” and “Security for us is priority No. 1”. These statements are mandatory, and everyone is very interested in security, because if you cannot ensure security, you cannot bring your product to the market. In this sense, security is a top priority. In fact, in almost every business, security certainly ranks third after functionality and price. We can list hundreds of product stories that could be safer if they cost more money or have less functionality. After all, unmanned vehicles that drive at only 10 mph,quite easy to make safe, but nobody needs them.

In fact, when car buyers are asked what factors they consider when choosing their next car, they always list “safety” as their first choice. Studies of what factors actually influence their choice have shown that in fact this factor is more likely in seventh place. Otherwise, no one would buy from anyone other than high-security brands such as Volvo and Mercedes, which at different times in history had the highest reputation in this area.

But Seltz-Axmacher points to a stronger factor, which is that the public, the press and investors are not interested in security, because it is boring by nature. And so it is. The perfect demo ride in a self-propelled car is boring, like a dishwasher. It’s hard to demonstrate security systems.

At the very beginning, when I was advising potential drone driving prizes to be on the heels of the DARPA competitions, I suggested holding a “man versus machine” safety competition. Vehicles would pass a difficult track, and fake obstacles, inflatable pedestrians and cars on small platforms would create problems. Experienced racers and robotic cars would compete in who could avoid a collision with various obstacles. This could attract attention - not when everything is perfect, but when any blows occur. When robots will ride perfectly, but the famous racers will not, it will be easier to gain trust among ordinary people. But nothing of the kind has ever been done, not even any attempt. That's because no one wants cars to appear on video,crashing into something (even if it's balloons), and also because it turns out that the task of automated driving is so complicated that handling various artificial situations has never become a priority. Now the teams are doing this in simulators. Sometimes tests are performed on test tracks, but they do not look exciting. (Waymo showed a video in which their car reacts to how employees throw boxes on the road).how employees throw boxes on the road).how employees throw boxes on the road).

Are all doomed?


I have not been to meetings of venture funds that refused Starsky more funding. The fervor of many venture capital funds declined, and many companies were refused. Perhaps they had other flaws that they did not want to talk about. I suspect that many companies will continue to receive financing, although some will suffer from the fact that initially they received high marks, which cannot be justified.

Perhaps the truth is that building robotic cars or trucks is not a game for small startups. It's quite complicated even for mega-startups like Zoox, Cruise and Aurora. A tremendous amount of hard and detailed work is required to increase performance from 99% to the required 99.9999%. The complexity of this transition is not equal to 1%, this work is 10,000 times harder, and not everyone understands this. The closer you are to great security, the harder and harder because it is more difficult to find problems, and each change can lead to problems in something that was fixed earlier. Perhaps we get a market only for big boys, at least for a few years. (It often seems that the things that billions went for the first time can also be done in the dorm room, after all).

Some companies are doomed to failure. A situation in which they all survived is simply impossible. This is expected when it comes to such a daring business. Big estimates require great results, and only a few companies can provide them.


image

About ITELMA
- automotive . 2500 , 650 .

, , . ( 30, ), -, -, - (DSP-) .

, . , , , . , automotive. , , .

Read more useful articles:

More articles:


All Articles