Understanding Autonomous/Self-Driving Cars

We've been dreaming about a self-driving car for years and it is depicted in many sci-fi movies as the final step in the evolution of the automobile. But how close are we to self-driving cars and will they be truly autonomous vehicles? To properly answer these questions, we need to understand the various aspects of autonomous driving. As we start the discussion, it would be prudent to get our definitions right. The Society of Automotive Engineers (SAE) does not refer to autonomous cars but prefers the word automated instead.

This is necessary to remind ourselves that self-driving car technology might be able to drive a car, but it cannot decide where we want to go. The human behind the steering wheel remains the truly autonomous piece of the puzzle, even in cars that drive themselves. Autonomous vehicle technology only takes over the driving part of the equation, or else we would live in a Matrix-type world where the machines have truly taken over - which is what full autonomy implies. So when we refer to fully autonomous cars in this article, it is with this proviso in mind, and because the term has become entrenched, even though it's technically inaccurate.

These are the five formally defined levels of autonomous driving:

• Level 0 - zero autonomy. The human driver has to maintain full manual control of the vehicle and performs all the driving functions, such as acceleration, braking, and steering. Such a vehicle might still be fitted with passive warning systems such as park assist.
• Level 1 - cars with driver-assist features. The vehicle has driver-assistance technology fitted that can intervene to control either the steering (such as lane-keep assist) or the speed (such as adaptive cruise control). However, only one can intervene at a time. These automated direction and speed controls help with some driving tasks.
• Level 2 - automation is partial. This is the first level where driving duties are shared between the driver and the vehicle, with the latter taking over accelerator and steering control, such as when the two assistance systems mentioned at Level 1 work together. Level 2 systems can even perform conditional emergency braking. The driver monitors conditions and is required to maintain full situational awareness. They are only allowed to remove their hands from the steering for brief moments.
• Level 3 - automation is conditional. At this level, some driving tasks become automated subject to conditions. The vehicle can monitor its surroundings, accelerate, brake, steer, and change lanes. Conditions may vary, but may be limited to certain speeds only, on certain highways only, or only in traffic jams. The driver is allowed to remove their hands from the steering wheel.
• Level 4 - automation is high. The vehicle can do everything it can at Level 3, but it can now also signal, turn, and handle unexpected driving conditions like obstructions, roadworks, or other complex driving situations without any input from the driver. The driver doesn't have to pay any attention to the road, nor do they need to have their hands on the steering wheel. However, they can take back control when they want to. The vehicle can also prompt them to take control but will bring itself to a safe stop if ignored. Level 4 is still limited by geofencing - where the functionality works in certain geographical areas only.
• Level 5 - fully autonomous driving. This is the truly automated vehicle. It can handle any driving task by itself with zero intervention by the driver, on any road, in any country, and under any conditions, with no restrictions or conditions. The driver becomes a passenger only. Needless to say, we aren't there yet.

So, how do self-driving cars work? Firstly, they need to be installed with the necessary equipment to detect their environment. This includes a myriad of sensors, processors, and actuators, run by smart control systems and machine-learning algorithms. Radar sensors and video cameras continuously map the entire area around the vehicle to identify the road surface and markings, pedestrians, traffic signs, and other traffic. Lidar (light detection and ranging) helps to identify road edges and accurately judge distances. The input of every single sensor is processed in real time by powerful software, which issues instructions to all the intelligent actuators to pilot the vehicle.

For a long time, everyone was stuck at Level 2 or somewhere between 2 and 3, which has been dubbed Level 2+. Even the Tesla Model S, with its much-vaunted FSD (full self-driving) feature could not actually break the barrier, and the Cadillac Escalade, which seems to have just about the best of everything in the world, didn't come close either.

Surprisingly, it was Honda that beat them all to the punch. Announced in November 2020, the Honda Legend was the very first prototype to boast true Level 3 autonomy. The car was officially launched in May 2021. However, considering all the obstacles that must still be overcome, Level 5 autonomy is a long way off.

We won't have a fully driverless car on the road until they can deal with the following, among many others:

• Traffic laws and conditions, such as tunnel-driving and carpool-lane protocols
• Differing driving laws - e.g. USA federal vs state laws
• Radar, lidar, and other sensors' cost and accuracy
• Navigation of all weather conditions successfully
• AI's ability to read human body language and react to their disregard for driving etiquette

While there have been studies that show car AIs are able to predict the behaviour of pedestrians by monitoring their body language, often better than people can, this same logic cannot be applied to drivers. A car drifting in its lane could be someone looking at their phone, falling asleep at the wheel, or switching lanes without indicating. A human might be able to 'feel' what is likely to happen, but AI struggles to make that determination, even with past experiences to draw from. This may be the largest hurdle engineers will need to overcome to develop truly driverless car technology.

Despite Elon Musk's bold claims, the AutoPilot and even the FSD features available on Tesla models are not true auto-driving cars, not by a long shot. Like Super Cruise, it's at Level 2 at most and there is still only one Level 3 vehicle commercially available. It will be a major step when the first Level 4-capable car hits the streets - where it's legal. When will self-driving cars be available to anyone walking into a car dealership? While we are, no doubt, moving in that direction, in terms of fully automatic driving, car technology still has a long way to go. Considering the changes needed and the required adjustments to road legislation, experts theorize that we should not expect to see autonomous vehicles until at least 2030.