How safe are robotaxis? BMW, Intel, Aptiv (and 8 others) just laid out a safety blueprint

Self-driving vehicles are often trumpeted as the answer to the millions of injuries and deaths that occur each year on the world’s roadways.

Developers and proponents of the technology argue that an automated system that can see, hear, react and make better decisions than humans will reduce the number of crashes that occur every year. And that’s a considerable promise. Some 1.35 million people died in vehicle crashes in 2016, according to the most recent statistics from the World Health Organization.

Still, there is no way for companies to guarantee that every self-driving vehicle put on the world’s roads can guarantee the same robust level of safety and security.

Eleven companies have formed a consortium to change that.

Tier 1 suppliers Aptiv and Continental, automakers Audi, BMW, Daimler, Fiat Chrysler and Volkswagen, as well as chipmakers Intel and Infineon, mapping consortium HERE and China’s internet search giant Baidu released Tuesday a 157-page white paper that outlines how to build, test and operate a safe automated vehicle. (The white paper is embedded below)

The idea behind the “Safety First for Automated Driving,” group (SaFAD for short) is to create a blueprint of sorts that lays out 12 principles for designing — and later testing and validating — safe automated vehicles, Karl Iagnemma, president of Aptiv Mobility told TechCrunch.

This isn’t meant to be a static effort. One technological breakthrough could render portions of the white paper moot. The intent was to create a “living” document that grows and adapts along with technology and the industry, Iagnemma said.

Nor does the document pick technological winners or losers. Those who sift through the white paper won’t find endorsements for certain sensors, for example. Instead, the group is pushing for common ground on safety.

The paper lays down certain “safety by design” rules that engineers should be thinking about and following from the beginning. This first section of the document lays out 12 guiding design principles, such as proper cybersecurity, protections if the system degrades or fails, operational design domain and data recording. A number of the principles deal with how a vehicle operator and the automated driving system interact with each other, including ensuring that if a handover does need to occur (this would be in vehicles that could switch between manual and automated controls) that it is properly and explicitly communicated to the human.

The principles also delve into the behavior of the automated vehicle to ensure that it’s predictable for the user and easy-to-understand for surrounding road users like bicyclists, pedestrians and other drivers.

The architects of this framework even weigh in on the safety of the interior of these automated vehicles. It’s a seemingly small, yet important guideline. So many of the futuristic concepts shown at tech and auto shows depict lounge-like interiors where people can face each other. The guidelines simply state that occupants should be protected even when there are new uses for the interior.

The second half of the document covers the validation and verification of these systems. In other words, how do you make sure that the automated vehicle you designed and built actually works? The consortium believes proper validation and verification should include testing on tracks, open roads and in simulation.

One more interesting note found within the pages of the white paper is an appendix that deals with machine learning and more specifically deep neural networks, namely how to ensure they’re developed, deployed and properly validated.

All of this, of course, sounds like common sense. And yet, companies have been working on automated driving systems without a single standard to guide them.

Many companies might be following similar rules laid out in the framework released Tuesday, but it’s impossible to know exactly what they’re all up to, especially when considering this on a global scale. The goal of the framework’s architects is to get as many companies as possible to sign on until they hit a critical mass within the industry.

Sure, companies have signed onto consortiums like this before to push for safe automated vehicles or promote public education campaigns. It’s worth noting that this effort was apparently driven by engineers within these companies, and not a marketing team or board of directors.

It should be noted that there is an existing international standard called ISO 26262 that covers the functional safety of electrical and electronic systems in production vehicles. And another standard, ISO 21448 or “Safety of the Intended Functionality,” is in the works to handle advanced driver assistance systems and autonomous vehicles. The SaFAD consortium as well as another group (the European Association of Automotive Suppliers, or CLEPA), which is being led by Nvidia, are involved in the development of ISO 21448.

The hope is that out of these various efforts, a single, clear standard rises to the top that can be validated and verified. Without it, companies may have trouble building trust with a wary and fickle public.