Man vs. machine – a software engineer and his Tesla
Meet Hans Christian Lønstad, CTO of Data Respons Solutions. A software engineer with 20+ years experience working at Data Respons, Hans Christian knows a thing or two about technology, and he is the proud owner of a Tesla Model 3. So, what would be more obvious than to ask him how that relationship is going – is that much-hyped car brand delivering on its promise? What are the upsides and downsides of owning a Tesla? And what are his thoughts on the current state of the automotive industry?
Published: 10. March 2021
By: Arne Vollertsen for Data Respons
Share
Hans Christian, why a Tesla?
– Initially I decided on buying an electric car to save money. In Norway electric cars are exempt from VAT, and they enjoy a number of other benefits. I studied the cars on offer and realised that Tesla is ahead of the competition when it comes to range and charging infrastructure, just to mention a few things. On top of that, everything works together seamlessly. A Tesla appears smarter than other cars.
What do you think of Tesla as a car manufacturer?
– I find it interesting to see what a com
pany can achieve, that has no history and is not bound by any kind of legacy. They started out with a blank sheet of paper.
– As I see it, Tesla is a great example of a truly disruptive business case, similar to when Apple launched the iPhone. Apple was the frontrunner, and afterwards all the Android phones produced in Asia followed.
– Now new car manufacturers are doing like Tesla. If I remember correctly, there are 10 new electric car brands emerging out of China. And just like Tesla they are starting from scratch. Also, the old car manufacturers are investing heavily in electric cars. Soon the electric car market will become fiercely competitive, and it’s hard to say if Tesla will be able to maintain its position. Who knows, maybe Tesla will continue as the flagship of electric cars. It’s still ahead of the competition, and that must be the reason why its value is so hysterically high, even though it hasn’t made much money yet.
What does owning a Tesla tell you about the current state of the automotive industry?
– If you look at a conventional car, it consists of a lot of subsystems, many of them manufactured by subcontractors. That concept worked well, as long as these components were isolated subsystems without the need for coherent communication and update mechanisms. But slowly everything became more and more dependent on communication between components (east/west) and to cloud platforms (north/south). A car may employ 25 or more computer modules and without a coherent software stack tying it together you’ll never be able to build a truly modern car.
What are the legacy car manufacturers doing to get past that barrier?
– They are investing heavily to develop a software stack and equipment configuration, and we’re already seeing some results. Volkswagen for instance is launching a series of electric cars based on the same platform. Some of the German software companies that are part of the Data Respons group are contributing to this new way of constructing a car, working for Audi, Mercedes, and others.
– But parts of the industry have had difficulties embracing that new approach to designing a car. About 4 or 5 years ago I attended a talk given by the head of development of Volvo. He told us that 70 per cent of development costs for a new model go into software, and only 30 per cent into mechanical components. That trend came as a shock to some vehicle industry executives, and now conventional manufacturers are investing enormous amounts in developing a state-of-the-art software stack and platform.
In your opinion, what can legacy carmakers learn from Tesla?
– An electric car is actually very simple. There are very few moving parts. Anyone could make an electric car. But the software required is where things get complicated. Here Tesla has a leading edge, and now other manufacturers are working hard to develop similar systems.
– However, I find at a bit strange that everybody is developing their own system. I wouldn’t be surprised if in 10 years time we’ll have an open source software stack that may be employed as a baseline for car manufacturers to license.
You’ve had your car for about a year. Are you satisfied with it?
– With the exception of the screen going black on occasion (you have a ctrl alt del on the steering wheel), the car has proven to be reliable. It also appears well built, although not quite on German premium cars standards.
– On top of that it’s fun to drive. My car is a performance model and it’s very powerful. You would have to pay 5 times the amount for a petrol car to get similar performance.
– And moreover, I’m looking forward to what Tesla has to offer when it comes to self-driving.
How so?
– I’m interested in how self-driving technology is developing. When I bought my car I paid a premium for the upcoming “Full Self-Driving” package, which is said to enable the car to find its way to a destination without any driver intervention. A beta version has already been distributed to a select number of Tesla owners in the US, but I’m not sure the package will even be available in Europe. We’ll see about that.
Are we going to see autonomous vehicles in the near future?
– I don’t think so. For years autonomous cars have been touted as the next big thing in automotive, but in reality many automakers are backing off on this promise and focusing on lower hanging fruits such as driver assistance systems. In general, I’m sceptical to the idea of full self-driving, at least when not confined to strictly regulated and provisioned environments. The general urban traffic scenario is highly complex and it is unlikely that machine learning can accomplish fluid traffic given the complexity of the task. Remember these algorithms depend on statistical confidence in order to make the right choice. Whenever a decision with potential safety implications is to be made, this confidence level must be very high – otherwise the car will probably have to halt. This is likely to be a recipe for a traffic jam.
– Tesla employs sensors as cameras, ultrasound and radar to establish the situational awareness required for autonomous driving. Cameras may however be blinded or confused by lack of contrast as is easily observed driving in winter conditions in Norway.
– There also exist ethic and legal aspects to the whole concept of autonomous cars. Who is responsible in the event of an accident? The non-driving driver or the car manufacturer? The algorithm supplier? What will the insurance cost become – if you ever get one?
How about your own car, how does it behave in traffic?
– A part of the screen will at all times show objects recognised by the car and thus give an indication of the situational awareness as perceived by the car. The deviation from my own understanding of the current traffic situation tells my something about the ability to navigate traffic – as a side note I have driven 40 years without accidents. In general, my experience is that the car definitely does not get the full picture. It is also dependent on lane markings to stay on track when auto steer is activated.
– One funny observation is that updates may lead to worse performance for instance in assisted breaking, probably due to stricter requirements to statistical confidence being put on the car makers. The car will generally appear more “nervous”.
Like all other Teslas your car is communicating with the Tesla headquarters. Have you noticed anything peculiar in that regard?
– For the car to receive updates and transmit data it has to connect to WiFi. In my house I’ve installed a sophisticated WiFi network, which allows me to see all clients and how much data they transmit. It tells me that when parked in my garage the car transmits considerable amounts of data after being taken for a drive. Its many sensors collect a lot of data, and Tesla is very good at using its fleet of cars to channel data back into their machine learning systems to improve them.
– That is one of the reasons why Tesla is ahead of the competition. They extract data to train their machine learning models. It is likely to be some kind of reinforcement learning, where they pick real world data related to situations when something unforeseen occurs. For instance, the car drives on autopilot and suddenly the driver grabs the wheel or steps on the brakes. I assume they want to analyse sensor data related to such an incident, and that makes good sense. But I’m guessing here, because the only thing I can see is that the car is uploading a lot of data. Exactly how it’s done, that is something Tesla keeps as a business secret.
Hans Christian, thank you for your time, and I hope you continue having fun with your car!