What needs to happen to make autonomous vehicles a reality?

Vehicle manufacturers and tech companies alike are pouring resources into autonomous vehicles but there remain huge technical, moral and legislative barriers that will be difficult to hurdle. It bodes the question; can fully autonomous ever truly happen?

Let’s start with some basic definitions. An autonomous car (also known as a driverless car, self-driving car, robotic car) is a vehicle that is capable of sensing its environment and navigating without human input. A classification system based on six different levels (ranging from fully manual to fully automated systems) was published in 2014 by SAE International, an automotive standardization body with 0 meaning ‘No Automation’ and 5 meaning ‘Fully Autonomous’ and requiring no human input. The most advanced cars on our roads today are at level 2 or ‘Partial Automation’. Cars that are part of trials and technology programmes are operating at level 3 or ‘Conditional Automation’ in controlled environments. The primary technologies involved include cameras, GPS and sensors; typically LIDAR to allow the vehicle to sense and understand its environment.

Companies including Uber, Apple and Tesla, as well as some of the major car manufacturers, are engaged in a race to develop vehicles that are capable of full automation. This holy grail is seen by some as an opportunity to radically change the personal transportation market from one of ‘private vehicle ownership’ to ‘rent a ride’. The basic premise is that vehicle ownership is sub-optimal for a number of reasons; it is a depreciating asset and it is massively under-utilised, spending most of its life on a driveway or company car park instead of in use. The thought of ‘apping a ride’ is alien to many of us suburban middle-agers but may prove very attractive to a younger generation of tech savvy city dwellers. If it happens then it has the potential to dramatically reduce the cost of personal transportation which will likely be the biggest driver for adoption.

But there are many obstacles to overcome.

Technology – A ‘Level 3’ car capable of conditional automation (still requires human supervision and input) has at least 300 million lines of code. The only way to build a system of that scale is through machine learning. It is envisaged that a level 5 fully autonomous vehicle will have upwards of 1 billion lines of code. It is no wonder that mathematics graduates with a post grad in computer science can name their price. Coupled with this challenge, the systems will need to be stable and reliable. Artificial intelligence isn’t currently capable of operating in chaotic environments. The various systems must function in challenging environments; cameras, for example, can’t see through heavy snow or low sunlight.

Legislation – International consensus will be required to build a framework of legislation for the use of autonomous vehicles which will inevitably result in differences of opinion. In a part- autonomous environment where does liability sit if there’s an accident? How do insurers even begin to measure the risk?

Moral judgement – Autonomous cars will not translate to zero accidents. The system will need to make moral judgements. Here’s a classic scenario; there’s about to be a head-on collision between two cars in an urban street. One is an autonomous vehicle . There are pedestrians close by. The autonomous vehicle has a split decision to make; either swerve and kill the pedestrians, most likely saving the occupants of both vehicles, or crash head-on with the other car, likely causing serious injury or death to the occupants of both vehicles. If we remove the pedestrians and replace them with a wooden fence then the car has a simpler decision to make doesn’t it? But what if there is a school playground immediately behind that fence? The level of awareness, knowledge and skill of the system will need to be at least as good as a human to work effectively and it is no mean feat.

From a resourcing perspective it is clear to us that the key players are genuinely confident that it can be achieved. Large car manufacturers like Ford are doing far more than hiring programmers; they are buying up entire software houses and partnering up with AR/VR gaming and Human:Machine Interface (HMI) specialist companies in a bid to stay at the bleeding edge of development. Tesla may be the company to watch as their approach has been radically different in that they have used their customers as the guinea pigs; the ‘Autopilot’ system installed in all of its new cars is constantly sending real world data back from the fleet operating in real world driving conditions all over the world back to the Tesla Mothership. This large scale ‘fleet learning’ may prove to be the only way of building a system big and complicated enough to achieve fully autonomous capabilities.