When Will Autonomous Cars Be Available?


The technologies for self-driving vehicles are maturing fast. Indeed, cars that drive themselves are operating on the streets of a city near you today, albeit with a safety driver inside, ready and available to take over. As the technical difficulties are significant, some believe that it will be a long and winding road to get to full autonomy. However, while this is true, the challenges are well-known, and the reality is that these are being overcome as we speak.

Some automakers claim that they already have complete self-driving capabilities, but these are limited to what is considered Level 2 autonomy (there are six levels – from zero, meaning no automation, to Level 5, which is complete automation).

Level 2 means that the driver still needs to be very much involved in driving. The leap to Level 3, where it’s completely hands-off unless the system “requests intervention”, is substantial, so for reasons of legal liability and to avoid potential litigation issues, many of the manufacturers don’t currently feel comfortable claiming full autonomy. When it does roll out, Level 3 will be a feature that can only be enabled at low speeds, such as acting as a chauffeur in a traffic jam.

This uncertainty means that in the short-term, car OEMs (manufacturers) are likely to follow the robotaxi approach to garner public acceptance of the underlying technologies – that is by running vehicles at slower speeds on known geo-fenced routes.

Of course, this may mean that it’s easier to get lorries driving in convoy on motorways and other highways on the basis that they can “platoon” and keep in contact with other vehicles ahead of them on the road, particularly if they’re driving known, well-travelled routes between depots.

These known circumstances lessen risk by making it more predictable, with fewer chances of variations or cases occurring that would be unknown to the vehicle. Humans are very good at learning from their previous experience to adapt to changing circumstances – but that takes years of practice in the real world. However, training neural networks so that they aren’t surprised by different eventualities and potential road incidentsis much more difficult.

Humans are very good at learning from their previous experience to adapt to changing circumstances.

Neural networks also need to be trained only on good driving practices – training an autonomous driving model based on humans whose driving is less than ideal (for example, braking late and skipping red lights) will result in this behaviour being copied by the self-driving algorithm.

Car manufactures will need to ensure they only learn from excellent drivers and follow ISO 26262 functional safety.

Take a look around you – and adapt

A self-driving car with multiple cameras and a 360-degree view around the vehicle can allow the autonomous vehicle to adapt to road users such as cyclists. It can see into the “human blind spots,” to give itself more room to navigate safely around the cyclist, whereas a human driver may not even have seen them. Safety then is of paramount importance on the road to production-ready autonomous vehicles.

Although the technologies exist today there are some concerns about unusual circumstances that the vehicle may not previously have encountered. A good example of an unusual circumstance that stumped early autonomous driving was in San Francisco, where mist or fog, and also steam rising from street vents, would confuse the algorithms, forcing the car to stop.

However, the capabilities of artificial intelligence are continuing to improve, thanks to extensive training in simulations and the thousands of hours of real-life driving experience that partially autonomous cars are gaining on the road. These sensor-enabled cars are absorbing vast amounts of data to continually train algorithms, thus enabling them to become more adaptable to these circumstances.

So while it’s not possible to put an exact date on the tech being ready, the date for autonomous cars becoming a reality remains tantalisingly close.