Most people who believe that autonomous cars will contribute to a more livable environment refer also at the potential savings of millions of lives. However, most statements about the safety of drivers, cyclists, pedestrians and not to forget animals in a world of driverless cars are speculative. One thing is sure, most people – 63% of all drivers in the US – say to be afraid of riding in self-driving vehicles. This percentage rose to 73% after two fatal collisions last year that received global attention.
To set the scene, a few words have to be spent at these accidents and their causes. On March 18, 2018, an Uber self-driving vehicle hit a woman who crossed the street with her bicycle. At that time, the mandatory ‘safety driver’ behind the wheel was watching the diagnostic instruments, which was allowed. Less than a second before the crash the driver looked at the road and managed to reduce speed. Analyses afterwards revealed that six seconds before the crash, the system identified the woman as ‘an unknown object’ and 4.7 seconds later the system wanted to engage the emergency brake, if it weren’t for the fact that this feature was at that moment disabled by the Uber team.
Less than one week afterwards, a Tesla Model X changed its direction without any necessity, hit a concrete barrier and caught fire, killing the driver. The driver had switched on the autopilot system on a road where its use was not authorized. He had also taken his hands from the wheel. This was also the case for another fatal accident that occured with a Tesla Model S.
The capabilities of autonomous cars differ to a very large extend. For this reason, the Society of Automotive Engineers (SAE) has developed a six-level classification (figure below).
Source: Society of Automotive Engineers (public domain)
Levels of automation
This classification discloses that cars with SAE level 2 can be trusted to steer and accelerate autonomously in an automatic mode in specific circumstances like motorways. Under these conditions, drivers can safely keep their hands from the wheel, provided authorization for that on those specific roads by national laws. As soon as the impact of the environment upon steering and accelerating is increasing, for instance after entering a busy street, the driver has to take over. SAE level 2 was deployed worldwide in 12 million vehicles in 2018.
All automotive and technology companies like Lyft, Uber and Google are currently working to qualify for the higher levels. Their expensive cars (up to $250,000) have the capability to monitorthe environment by the simultaneous deployment of cameras, radar, Lidar and high definition maps. A properly working SAE level 3 system enables drivers to keep their eyes from the road and to be engaged in any other activity. The only condition is that they must be ready to take over driving immediately as soon as the system gives the disengagement signal, meaning that is in unable to handle the situation.
For prospective driverless cars offering cheap taxi services, this level of mastery is not satisfactory. At SAE level 4, cars have automated fallback performance, which means that they can handle any situation under specified circumstances, like well-designed roads, during day-time and at a certain speed. SAE level 5 automatization includes the ability to proceed driverless under all circumstances. Not any of the existing autonomous models is complying with these requirements.
A prototype of an autonomous SAE level 5 car deluxe. Photo: Mercedes
Although SAE level 5 is the ultimate ambition, the automotive industry is in no hurry to remove the driver behind the wheel. Their mission for the next decade is selling as many as possible electric cars to private persons, with (semi-)automatic systems as nice-to-have features. Therefore, consolidating SAE level 3 is their first priority.
At the other hand, technological companies like Google, Lyft and Uber cannot wait acquiring SAE level 4, which opens the way to driverless taxi services.
Every year, Navigant Research ranks automotive and other companies trying to develop self-driving vehicles (see table below). In 2017, Waymo, a subsidiary of Alphabet, and Cruise, a division of GM, are in a leading position, while Apple, Uber, and Tesla are staying behind, which might explain Ubers irresponsible testing-behavior.
The frequency of disengagement signals during the first part of 2017 is a clear illustration of the difference between these companies. In average, Waymo cars drove 9000 km before the ‘safety driver’ had to intervene. Cruise’s drivers had to be a bit more attentive and had to take over once during 2000 km. Uber drivers were alerted every 20 km. Both Waymo and Cruise refer to still much better results in 2018, increasing the gap with Uber.
Ranking automotive and technology companies. Source: Navigant Research (public domain)
Waymo is in the process of rolling out a SAE level 4 autonomous taxi service in a suburb of Phoenix (Arizona) and GM has announced similar plans for San Francisco in 2019. Uber’s bad performance is strategically much worse than Tesla’s. Uber needs a SAE level 4 classification to introduce a huge number of cheap taxi’s. Egon Musk’s ambitions are higher than the sky, but the lions’ share of his customers fall for the gorgeous driving capabilities of the Tesla and are not waiting to let ‘the system’ do all the work.
Back to safety issues
Automotive and technological companies have been testing autonomous cars at length. Waymo alone counts 9 million kilometers without severe accidents and in no accident the driverless car was in fault. However, all of those kilometers were driven with a safety driver on board, being able to deal with any disengagement signal and to prevent other evil.
If the accidents in which Uber and Tesla have been involved, disclose anything about safety, it is that Uber’s Volvo cars definitely are not ready yet for autonomous driving at SAE level 4, apart from the irresponsible behavior of the team. Considering Tesla, it is that the self-driving qualities of its cars are less superb than Egon Musk wants us to believe (at highest at SAE Level 2) and that overreliance on the TESLA autopilot is deadly dangerous.
From the perspective of road regulations, automated vehicles (SAE Level 3, with a human driver) and autonomous vehicles (SAE Level 4 and above, without a driver) differ in a fundamental way. Regulations regarding autonomous vehicles have prevented driverless cars to appear anywhere until now, although no serious accidents happened so far during their testing. The accidents with SAE level 2 cars mentioned above raise questions about the unambiguity of the conditions under which automated systems are allowed or prohibited.
The State of California recently proposed new rules to enable driverless cars. Several other states in the USA followed. As a consequence, Google and General Motors have been authorized to launch driverless taxis services shortly, albeit with a mandatory presence of a safety driver in the pilot period. The trips will be monitored using cameras to avoid reckless behavior or vandalism.
The advanced system of law making in California has been exemplary for global law making with respect to autonomous cars in the future. The conditioned permission to deploy driverless cars in certain areas illustrates the fact that the safety of autonomous cars is increasing, but also that many safety issues still have to be solved. Driverless cars struggle with ‘randomness’, caused by children, pedestrians, cyclist and human drivers and also with potholes, detours, and worn markings. Traffic lights pose a problem too. After all, the whole world has been able to watch an Uber car driving through the red light.
A more detailed video can be found on website of The Guardian.
Until now, and in the (near?) future the construction of driverless cars is based on safe participation in traffic that is dominated by ‘ordinary’ cars and other users of the road. Real progress in safety will be reached once driverless cars are enabled to communicate with each other and the presence of human-driven cars on public roads is prohibited, not to speak of horse-drawn vehicles. Safety will thrive if the urban environment is reshaped in areas where autonomous cars dominate and areas for cyclist and pedestrians. In this case, the focus of the discussion is shifting from the merits of autonomous cars to the quality of our living environment.
*) Source header: Waymo
**) This article was brought to you by Professor Herman van den Bosch, Professor at Open University of The Netherlands.