Self-driving vehicles may already be a reality, writes John Frazer, CCO, DAV, but the underlying systems that will make them work are still being improved and fine-tuned. What are the challenges and what can we realistically do about them?
Building a self-driving car takes more than strapping sensors and software onto a set of wheels. Let’s start with the obvious: autonomous vehicles need to navigate not only other travelling automobiles, but objects that may be on or even in the road. This involves avoiding large, inanimate objects but also jaywalkers.
More challengingly, it means discerning between smaller objects and their potential to cause harm to the car and its passengers – let’s say, a plastic bag caught by the wind, and a small deer. This requires an exceptional level of intuition that is difficult to program for, because size doesn’t necessarily correspond to the level of danger posed. Weight, shape, brittleness and malleability all come into the equation.
The solution involves both adapting the environment outdoors and improving the car’s intelligence: on the one hand, designing roads which, to the fullest possible degree, provide clear visual cues to cars, and on the other, designing artificial intelligence that can decipher complex visual cues with speed and precision.
Of course, there will be times when the road environment isn’t as clear as it should be. Cars need to drive safely regardless of whether their paths are demarcated with white lines or cat’s eyes. Moreover, in the absence of functional traffic lights, autonomous vehicles must continue to make correct decisions.
For objects that are simply difficult to gauge, AI will need to err on the side of caution, especially as the technology is in its infancy. Vehicles need to be programmed to seek a safe state on their own, such as pulling to the side of the road and stopping without relying on the driver for manual input. False positives will undoubtedly arise, and unnecessary stops will be made, but that’s the price of safety.
Communicating in real time, with everything
The other major issue facing autonomous vehicles is communication. How do these vehicles function in an entirely self-sufficient ecosystem – that is, one in which autonomous vehicles almost exclusively make use of other autonomous vehicles? How does one autonomous vehicle pay another for a service?
As things stand, there is no infrastructure in place to facilitate all this. Uber, Google and Waymo are all trialling self-driving vehicles that may well be on our roads in the future – but there isn’t any connective tissue linking them.
Tech giants are building their solutions using the same closed platform model, and while smaller firms are emerging, their networks are equally proprietary, closed and non-inclusive. The answer is most likely to be blockchain, which would allow vehicles to discover each other and make use of other autonomous vehicles and services.
On the other side of the coin, there’s also human communication to consider – both inside and outside the car. On the inside, if level three automation is reached – an imperfect level of automation – car manufacturers will need to consider how to get the attention of the driver if the automated systems fail. Given factors such as sleep, stress, inattention and anxiety in humans, designing the process for a human taking the wheel isn’t easy.
On the outside, developers of sensors need to figure out how to manage all the subtle behavioural signals that humans take for granted: the body language of a police officer, for example, as they signal for your car to pull in, or a cyclist putting out a hand to indicate a turn. This presents even less of an obvious answer. Perhaps it’s a matter of endless hours of road testing, or perhaps as far as human body language is concerned, it’s well above our technological capabilities, at the moment.
A brave new autonomous world
There isn’t a unified strategy to rectify the problems outlined above, but one thing’s for sure: to make rapid progress, we need to cooperate. At the moment, large businesses with stakes in autonomous vehicles are naturally focusing on proprietary technology and dominating their own markets, rather than investing in tech to benefit all, including competitors. But that’s not how significant headway will be made.
The autonomous landscape of the future involves everyone – businesses, government and citizens – coming together to invest in and develop open-source technological and real-world infrastructure for vehicles to rely on. If the collaboration doesn’t happen, there’s a danger that we could see little growth in autonomous vehicles over the coming years, with no real improvement in safety, congestion and the environment.
See DAV for more information.
The author of this blog is John Frazer, CCO, DAV