we.CONECT: What are your main responsibilities in your current role?
László Kishonti: As the founder and CEO of AImotive my main goal in the current economic climate is ensuring the long-term success of the company and providing the team with the best resources and the highest level of security possible for them to continue their outstanding development work efficiently. I work closely with our product teams to ensure that our developments are in line with the development goals we have laid out, and with our commercial organization to grow our partner base.
we.CONECT: How is it like working in Autonomous Driving R&D in your company today and on which extend do you focus on future level 4 & 5 autonomous driving technologies?
László Kishonti: Three years ago, everyone believed that self-driving cars would be on roads by 2020 or 2021. Today we know this won’t happen, so the focus has shifted to Lv. 2 and 2+ applications which are already advanced enough to save thousands of lives. Naturally, the long-term goal is still Lv. 4–5 but the next decades will bring a gradual evolution of technology rather than the sudden revolution everyone expected a few years ago.
This change of pace, and the current situation of the automotive industry, and the crisis caused by COVID-19 will lead to industry consolidation. Only companies with meaningful projects and real-world solutions will survive. In the long run though, the key will be finding a balance between development speed and safety. This is where AImotive helps its partners most.
we.CONECT: How big a role do you see current & hybrid AI-Approaches and cognitive computing, such as self-supervised & behavioral learning, playing in self-driving car technologies?
László Kishonti: Current automated driving systems rely heavily on what are known as self-supervised and behavioral algorithms. When in operation these solutions gather a vast amount of data, which can then be used to further develop and finetune the solution, leading to over-the-air software updates and a better user experience.
The result? Software is the new king of the automotive industry: Tesla is currently the most valuable OEM in the world in terms of market capitalization, while it manufactures a fraction of the cars made by larger traditional OEMs. The difference in perceived value comes from the software, and Tesla’s ability to improve the user experience over time releasing over-the-air updates and running new functions on the existing hardware. This means the vehicles maintain their value and feature roll-out time is faster. Even consumers are subscribing to the new approach. A recent survey showed that Tesla remained extremely popular and highly valued despite being one of the least reliable OEMs when considering mechanical failures. Traditional OEMs will have to adapt to this new view of the industry if they want to survive.
Artificial intelligence (AI) is a core element of this software technology as it solves problems that cannot be overcome through traditional algorithms. These are complex questions such as predicting driver intent. This is something human drivers do constantly by reading other vehicles in traffic and the environment – but is extremely complex and based on earlier experience and presumptions. Traditionally, computer systems have tried to do this by first detecting vehicles as bounding boxes and then the algorithm used the motion of the bounding box and a dynamic model of the vehicle to make predictions. This method creates a bottleneck since the algorithm is not able to utilize a subtle clue hidden in the raw sensor data, such as a pothole leading to an evasive maneuver. In contrast, AI can be trained for these clues with the help of self-supervised learning and behavioral learning which is a significant advantage.
we.CONECT: To what extent has the automotive industry achieved a sufficient level of perception & prediction accuracy in sensor suites to enable operational safety for hard-to-predict road actors and in adverse weather such as very heavy rain, falling snow and fog?
László Kishonti: Changing weather conditions will affect the level of the solution offered to drivers when travelling. For example, on a highway under clear skies the vehicle may be able to handle almost all driving tasks, however, as conditions worsen and for example, fog moves in, the system will increase its reliance on the driver. At the current stage of the technology, in the worst conditions the software will simply hand back control, or better yet, not even take control, of the vehicle to the driver in a controlled setting. Diverse sensor modalities (such as cameras, radars etc.) can expand the range in which the car can operate. However, in poor conditions, with reduced visibility, human drivers also slow down as the reaction time changes – so it’s no surprise that this means the same adaptations for an automated driving system. Furthermore, from a safety perspective the most important thing is or the vehicle to recognize that it cannot see clearly.
To solve level 5 automation infrastructure will also have to develop. Horse-drawn carriages for example could use dirt roads easily, as they never reached speeds that made these truly dangerous. However, when cars came along and speeds increased more and more roads were paved. Automated driving will see a similar shift, with road infrastructure increasingly been resigned and rolled out to help automated systems navigate safely. Support from infrastructure is one-way level five vehicles may be able to handle all weather conditions in the future.
we.CONECT: How did the global Covid-19 outbreak change your personal view on autonomous driving?
László Kishonti: In the short term, of course, the outbreak caused the entire automotive and autonomous driving industry to decline across the globe. In the mid-term, as I said earlier it will lead to consolidation and only the fittest companies will survive. Automotive OEMs have already realized that the market is changing rapidly, and with the budget restrictions caused by the severe hit on the industry only a few of them will be able to handle the whole R&D process in-house. The others will soon turn to available, accessible and deployable solutions, and that is where AImotive can offer a helping hand – and with this we can not only survive the crisis but come out of it stronger than we were before.
we.CONECT: What would you change in the AV industry to speed up and improve autonomous R&D in the post-pandemic era?
László Kishonti: I think there are three main factors at play in the AV industry at the moment. The first is the industry consolidation that began before the pandemic, but will no doubt be accelerated by it. The AV industry is extremely fragmented with too many small players battling to survive in small areas, take a look at the number of LiDAR companies. Consolidation will clear the waters, leaving only those companies standing that have meaningful solutions to real world problems. This new focus will accelerate development in the long run.
The role of simulation is also essential. Simulation is the only solution that can ensure quick development without sacrificing on safety. Drawing inspiration from the aviation industry we began working on a purpose-built simulator when barely anyone was using them in automated driving. The result? AImotive has created the only ISO26262 ASIL-D certified automated driving simulator in the world, enabling our partners to accelerate their development drastically. But simulation is only half the solution, the number of tests to be created and run, the data to be collected, and processed is too large for any human team to handle. Having a robust automation pipeline, cloud-based solutions and meaningful reports is key to progress.
For me, it is completely unthinkable that 10 years after the release of the iPhone and five years after Tesla software updates started to be released it is still not a standard feature for 99% of brand-new cars to receive software updates. This is the feature that is fundamentally capable of connecting consumers to the car as it is constantly improving it. It also shows just how much the automotive industry is not software focused. That should change, and this is the area where AImotive can help.
we.CONECT: What are your major predictions and key future challenges for autonomous driving?
László Kishonti: The main challenge OEMs face is how to catch Tesla or as I said earlier, when will the automotive industry focus on what’s really important: software. Traditional OEMs are usually hardware focused and may lack the software expertise needed to respond to these changes. This is where companies like AImotive come in. Smaller, independent and fast moving with a deep understanding of both software development and the automotive industry, our goal is to provide solutions that our partners can use to advance their own development efforts. No one is going to solve automated driving alone, collaboration will be key in this race, that is why all of our technologies are hardware agnostic, modular and easy to integrate into existing systems. The other interesting factor is that cars can still contain more than 100 small computers (ECUs) – this is unthinkable in any other consumer industry, and it is the main obstacle for software updates, because you can’t update 100+ computers at the same time efficiently. My prediction is that the number of compute units in a car will drastically decline.
The other major shift will be when vehicles become platforms, rather than end products. A similar change happened when the smartphone was born. Originally, when a user bought a phone, its functionality stayed unchanged for much of its lifetime. When the smart phone appeared, this changed. The phone’s functionality could be expanded through over-the-air updates and, even more importantly, apps (and app stores) appeared providing even more value to users, and new revenue streams to manufacturers. I think the automotive industry will experience a similar shift in the coming years, and OEMs will have to capitalize on this as much as they can.
we.CONECT: What are the key aspects of your session at the Auto.AI Europe 2020?
László Kishonti: I will be focusing on some of the changes coming to the automotive industry, and how traditional OEMs can answer the challenge Tesla has placed before them. I’ll go into more detail on some of the points I touched upon above. For example, the factors that drive Tesla’s market capitalization, for example: how the market, and even consumers, are placing a larger emphasis on software than mechanical reliability. I will also take a look at the importance of simulation and automation in developing safe automated driving systems while accelerating feature roll-out times.
&nsbp;
