For more than a decade, car manufacturers have been working on technology to take over driver’s actions. A Lot of money has been invested in this short period and many optimistic expectations have been raised, but no large-scale implementation of the higher SAE levels resulted so far. Commercial services with robotaxi’s are scarce and still experimental.
The changing tide
Especially in the period 2015 – 2018, the CEOs of the companies involved cheered about the prospects; soon after, sentiment changed. In November 2018, Waymo CEO John Krafcik said that the spread of autonomous cars is still decades away and that driving under poor circumstances and in overcrowded cities will always require a human driver. Volkswagen’s CEO said fully self-driving cars “may never” hit public roads.
The companies involved are therefore increasingly concerned about the return on the $100 billion invested in the development of car automation until the end of 2021. The end of the development process is not yet in sight. Much has been achieved, but the last 20% of the journey to the fully autonomous car will require the most effort and much more investment. Current technology is difficult to perfect. “Creating self-driving robotaxi is harder than putting a man on the moon,” said Jim Farley, CEO of Ford, after terminating Argo, the joint venture with Volkswagen, after the company had invested $100 million in it.
The human brain can assess complex situations on the road much better than any machine. Artificial intelligence is much faster, but its accuracy and adaptability still leave much to be desired. Driverless cars struggle with unpredictability caused by children, pedestrians, cyclists, and other human-driven cars as well as with potholes, detours, worn markings, snow, rain, fog, darkness and so on. This is also the opinion of Gabriel Seiberth, CEO of the German computer company Accenture, and he advises the automotive industry to focus on what is possible. Carlo van de Weijer, director of Artificial Intelligence at TU Eindhoven, agrees: “There will not be a car that completely takes over all our tasks.”
Elon Musk, on the other hand, predicted that by 2020 all Tesla’s will have SEA level 5 thanks to the new Full Self Driving Chip. In 2023 we know that its performance is indeed impressive. Tesla may therefore be the first car to be accredited at SAE level 3. That is not yet SAE level 5. The question is whether Elon Musk minds that much!
The priorities of the automotive industry
For established automotive companies, the priority is to sell as many cars as possible and not to make a driver redundant. The main objective is therefore to achieve SAE levels 2 and possibly 3. The built-in functions such as automatic lane changing, keeping distance, and passing will contribute to the safe use of cars, if drivers learn to use them properly. Research shows that drivers are willing to pay an average of around $2,500 for these amenities. That is different from the $15,000 that the beta version of Tesla’s Full Self Driving system costs.
The automotive industry is in a phase of adjusting expectations, temporizing investments, downsizing involved business units, and looking for partnerships. GM and Honda are collaborating on battery development; BMW, Volkswagen and Daimler are in talks to share R&D efforts for autonomous vehicles; and Ford and VW have stopped developing an autonomous car and are working together on more realistic ambitions.
Safety issues at SAE level 3
But even with a focus on SAE level 3, the problems do not go away. The biggest safety problem may well lie at this level. Elon Musk has suggested for years that Tesla’s autopilot would allow drivers to read a book or watch a movie. All they must do is stay behind the wheel. They must be able to take control of the car if the automatic system indicates that it can no longer handle the situation. Studies in test environments show that in this case the reaction time of drivers is far too long to prevent disaster. An eye on the road and a hand on the wheel is still mandatory everywhere in the world, except in few paces for cars accredited at SEA level 4 under specified conditions.
The assumption is that the operating system is so accurate that it indicates in time that it considers the situation too complex. But there are still many doubts as to whether these systems themselves are sufficiently capable of properly assessing the situation on the road at all times. Recent research from King’s College London showed that pedestrian detection systems are 20% more accurate when dealing with white adults than when dealing with children and 7.5% more accurate when dealing with white people compared to people with dark skin.
In the next post I will go into more detail about the legislation and what the future may bring.
Music: Listen, Look and Enjoy 