One can also teach self-driving cars how to follow the highway code. This necessitates dissecting each rule in the code, teaching vehicles' neural networks how to obey each rule, and then verifying that they can be trusted to follow these rules safely in all situations.

FREMONT, CA :Autonomous vehicles, ranging from warships to supermarket delivery carts, have recently made a smooth transition from testing labs to the real world with few hiccups. Self-driving cars, on the other hand, have yet to make it through the testing stage. In addition, a pedestrian was killed by one ride-hailing company's self-driving car during a test drive in 2018. While human-driven cars are involved in incidents daily, the public holds driverless cars to much higher safety standards, viewing single accidents as evidence that these vehicles are too dangerous to be released on public roads.

It is a big and technological challenge to program the ideal self-driving car that will always make the safest decision. Unlike other autonomous vehicles, which are typically tested in managed environments, self-driving cars must operate in an endlessly volatile road network, quickly processing several dynamic variables to stay secure.

Stay ahead of the industry with exclusive feature stories on the top companies, expert insights and the latest news delivered straight to your inbox. Subscribe today.

Highway coding

To keep all road users safe, human drivers follow a highway code that relies on the human brain to learn and apply these laws in various real-world scenarios. One can also teach self-driving cars how to follow the highway code. This necessitates dissecting each rule in the code, teaching vehicles' neural networks how to obey each rule, and then verifying that they can be trusted to follow these rules safely in all situations.

When looking at the ramifications of the word ‘must never’ in the highway code, ensuring that these laws can be followed safely becomes more difficult. To make a self-driving car respond as quickly as a human driver in any given situation, we must program these policies to account for complexity, weighted risk, and the rare case where several rules are in direct conflict, causing the car to disregard one or more of them.This is not a job for programmers alone; it will take feedback from attorneys, security experts, system engineers, and policymakers as well. Teaching self-driving cars to be perfect will be a fluid process that will change over time as legal, cultural, and technical experts define perfection.