The automotive industry has accepted that full scale autonomous driving across the whole road network under all conditions in dedicated vehicles without any steering vehicles or other manual controls will not happen until 2035 and probably later. But intermediate levels of self-driving will be introduced much earlier, able to exploit some of the potential cost savings and safety benefits of conditional autonomy.
Governments and regulators are welcoming some conditional autonomy under restricted conditions, which will help push improvements and provide the data necessary for subsequent expansion.
Even during the ongoing Covid-19 coronavirus pandemic, this is bringing back some optimism to a field that had become somewhat deflated and derailed by a spate of accidents during trials, mostly in the USA. It was only last month in September 2020 that Rafael Vasquez, the back-up driver of an Uber self-driving car that killed a pedestrian during 2018, was charged with negligent homicide. Earlier investigations by police and the US National Transportation Safety Board (NTSB) found that human error was mostly to blame, but the point was it highlighted the state of limbo that occurs during the so-called Level 3 transitional stage of self-driving.
Although the vehicle can drive on its own under many conditions under Level 3, a human back-up driver has to be on hand to take over almost immediately, while in this case the accused had been found streaming an episode of the TV show ‘The Voice’ while he was supposed to be fully alert by the wheel ready to take over.
While Uber itself escaped criminal charges, the State of Arizona where the accident took place suspended the firm’s license to test self-driving cars on the state’s public roads. Uber postponed its autonomous programme, while Tesla was also under fire for two fatalities involving its self-driving cars.
These accidents provoked a rethink of the whole autonomous driving roadmap among both automotive manufacturers and regulators. The idea had been that self-driving would be introduced progressively in the six-level model that enjoyed universal support around the world. Under this, Level 0 is zero autonomy, Level 1 introduces some of the sensory features required by higher levels, such as collision detectors, but confined just to issuing bleeping warnings at this stage, as when parking. There might also be adaptive cruise control to limit distance from the vehicle in front and set the maximum speed.
Level 2 then introduces partial automation so that drivers can temporarily hand over some driving functions such as steering, acceleration, braking, speed maintenance and lane centering. However, these will only work when drivers maintain both hands on the wheel so that they are ready to take back control if necessary.
This then sets the stage for Level 3, which is where the problems can start, because this is the first level that can be considered autonomous in any real sense. For that reason, and as a result of those accidents, even plans for Level 3 driving had been put back, on the part of manufacturers, regulators and also hiring firms or fleet operators such as Uber.
Just two manufacturers, Honda and Audi, came forward with full designs for Level 3 cars, while Japan has stood out by encouraging Level 3 driving, motivated by the country’s desire to show off its technology prowess at the 2020 Tokyo Olympic Games. That of course was postponed for a year because of the pandemic, but not before the country in June 2019 had amended its Road Traffic Act to allow Level 3 automated vehicles to be used on its public roads from May 2020.
This was under certain conditions, such as having vehicles fitted with a data recorder analogous to the black box recorders of civil aircraft, and requirement for approvals through a permit system to control wireless updates to the self-driving software, to minimize risk of cybersecurity breaches.
The ambivalence of Level 3 was highlighted by drivers being allowed to watch TV or read a book while the autonomous system is in charge, and yet be fully vigilant to take over if some exception arises. There could be insufficient time for the river to respond in the event of an impending accident.
For this reason, most carmakers have been reluctant to bring out Level 3 cars, with some such as Volvo now planning to skip straight to Level 4 on the basis that this is clearer cut. Under Level 4, drivers can hand over control completely without being ready to resume, so can sleep if desired. The only limitations are that certain locations or situations may be prohibited, perhaps the fast lane of trunk highways or motorways, and the vehicles must still have full controls so that they can be driven manually if the autonomous systems fail completely. Level 5 then allows full autonomy everywhere in redesigned dedicated vehicles without steering wheels or other controls.
So that leaves Honda now as the only automotive OEM still planning to introduce a Level 3 car this year, just in Japan, Audi having delayed its planned launch. This leaves Level 3 virtually dead in the water, but meanwhile it looks like arising from the ashes as what looks like a limited version of Level 4. The UK could take an initiative here since its government is considering allowing automated technology to be trialled on some of its motorways under limited conditions later in 2021.
The aim is to exploit functions already present in many non-autonomous cars, primarily collision avoidance, speed control and lane-keeping. The idea is to progress from provision of alerts and driver assistance, which happens anyway, towards more automated control including steering.
The UK government had noted an EU report finding that 95% of all road accidents across member states, including 25,300 deaths in 2017, are caused by human error.
Already autonomous features have cut this toll somewhat just through provision of alerts and guidance, but the UK thinking is that more substantial inroads into the death toll can be made by automating collision detection, lane steering and control of distance from the vehicle in front. This could be seen less as a rethink of Level 3, more a progression towards Level 4 under more constrained conditions. Then eventually the scope for such autonomy would be increased progressively.
The other area where self-driving looks like being introduced earlier than elsewhere is in platooning, primarily of commercial good vehicles. The idea here is that the front vehicle has a human driver as normal, while following vehicles close behind, initially confined to one, are self-driven. Again, to start with there would be a human safety driver ready to take over.
Platooning trials have been underway for several years and the field has slowly been advancing into more open situations, so that late in 2019 for example agricultural cooperative Land O’Lakes in the US was using an autonomous truck from Plus.ai to test coast-to-coast deliveries of its butter.
Another pioneer is Wilson Logistics, a trucking company operating across the whole contiguous US, working with vehicle technology platform company Locomotion to add autonomous relay convoy technology to its trucks. This again has one driver in the lead truck followed by several others in tandem. Wilson reckons it will cut operating costs by 33% and fuel costs by 8% across a fleet of 124 trucks initially.
Platooning will be an important proving ground for more advanced autonomous technology because it is a controlled environment where human drivers will still be in charge at the front, with the following vehicles enjoying close coupling for collision avoidance. Over time vehicles will be able to exploit roadside 5G-based C-V2X communication as well, which could for example help anticipate hazards further ahead beyond the sight even of the lead driver.
Another area where autonomous driving technology could cut its teeth earlier is remote driving, where humans retain oversight but from a centralized control room. This is really being ushered in by 5G, given the obvious need for wireless communications at guaranteed low latency and with high availability. This operates in tandem with self-driving, which in a sense provides a back-up if communications did fail even briefly. As well as being seen as an interim stage towards eventual full autonomy, this offers scope for savings since a given team of drivers might be able to control a larger number of vehicles.
Initially, trials will be confined to vehicles such as buses operating on well-defined limited routes, as in a two-week pilot running in Stockholm, Sweden, from September 24th to October 8 2020. This involved transportation company Keolis, in partnership with a smart city platform called Urban ICT Arena, telco Telia, Ericsson, Intel and vehicle control system maker T-engineering. This featured a self-driving, electric minibus in Stockholm remotely controlled over 5G, exploring how to introduce autonomous electric vehicles in complex urban areas, with a focus also on route planning and traffic control.
It is easy to imagine remote control being applied successfully in conjunction with autonomous driving in campus and industrial environments with UAVs (unattended automated vehicles). Overall, the autonomous driving field will advance through stealth by establishing safety and proficiency in these more constrained environments, before eventually being unleashed across larger public road networks.