Zenseact logo neg

Better safe than sorry

Some of us find it hard to trust a machine to take control of something as important as driving. What if the almighty AI fails? It’s still a relatively new technology, and some of us might be uncomfortable with the idea of not having a human driver at the wheel.

Moreover, self-driving cars could be vulnerable to hacking or other cybersecurity threats, which could compromise the safety of the passengers or lead to other consequences. And what about all that data being recorded by cars driving around? Who controls the use of that information? Is it anonymized (spoiler alert: yes, the data we collect is thoroughly anonymized). As self-driving vehicles become more prevalent, there will be new questions and challenges related to cyber security, data, personal integrity, liability, insurance, regulations, etc.

And honestly, many of us genuinely enjoy the act of driving and the sense of control and freedom it provides. So why would we give up this pleasure and autonomy to a machine?

And we probably won’t, not entirely. Currently, full automation is more of a robotaxi thing. For passenger cars, it will likely be more about developing a nice, smooth relationship with automation, even if it’s anybody’s guess what people will do with their newfound freedom behind the wheel.

Regardless, as self-driving technology continues to develop and prove its safety and efficiency, it’s likely that skeptics will gradually become comfortable with the idea – just like it took years, decades, of safety improvements for us to accept riding in elevators or getting into airplanes. The shift from exclusive indulgence for the few to secure widespread adoption requires time.

Now, to convince the public that our automation is safe, we can outline our belts and brace approach to developing it. We can try to convince people that we do not compromise when it comes to safety.

For one, there’s the legal aspect we take into consideration. This is when AD systems demonstrate their safety by showing they comply with standards, certifications, regulations. And this is important; without standardization, we cannot ensure a given product’s quality, safety, and inter-compatibility. Signing such a societal contract also makes AD development move forward on a unified front, gradually instilling more confidence in the public.

Some will argue that this is enough. Design a system that complies with the rules, and you’re good. But this standards-based approach has its flaws. Even if you fulfill the criteria, can you really be sure your exceedingly complex end goal is met? What’s your level of ambition, your mission, your purpose as a company?

Achieving what you have set out to do is something you might have to take a lot of responsibility for yourself. That’s just how it is.

Put differently, a compliance-based approach would be enough in a perfect world where everyone did as they were told. As they were trained to do. To be sure, a strictly “rule-based” approach makes it easier (at least hypothetically) to determine whose fault is in a crash. However, an “It wasn’t me!” approach won’t prevent the accident from happening. It’s pointless in the bigger picture, where any accident should be avoided, not just the ones you would be blamed for.

We’ve touched on the fact that people often break traffic rules. They speed, get distracted, misinterpret signs, run through intersections, and so on. Generally, we’d view these actions as improper, yet ironically, they can sometimes help the traffic system function. For instance, wouldn’t we consider nudging past a red light by a meter or two, if it meant making way for an ambulance, provided it was safe to do so?”

In a sense, then, the game is rigged: how successful can a risk-averse robot following robot rules really be in a world of “flexible” and experienced humans?

At Zenseact, we’re convinced that developing technology that “follows the rules” and considers human behavior is the safest bet. It shouldn’t matter whose fault the accident is. And it won’t if we can avoid the accident altogether.

It’s a little bit like this: If you crash into a tree, what’s the lesson learned? Figure out how to not drive into trees? Remove all trees close to the road? Plan to drive only where you can see the road clearly? We suggest doing all of the above, leaving you prepared in case there’s a tree left behind that suddenly decides to fall over the road. The key takeaway? You need a comprehensive approach to automation that incorporates both proactive and reactive measures.

The core idea of our approach to safety is to drive cautiously, enabling collision avoidance and permitting emergency maneuvers in the event of unforeseen occurrences.

Now, this was safety in theory. At some point, you must also show that your solutions work. Following good old evidence-based practice, you might share collision avoidance statistics and the number of hours driven without accidents or severe injuries. You can demonstrate how you’ve reduced the number of emergency interventions by drivers using the software. This will give an indication of how well the software performs. But it’s not enough. Such data isn’t available until years after the technology is introduced, and what good is it then for assuring someone that the technology is safe from day one?

What’s more, even if one would come up with great statistics and an impressive track record, it would still be “so far, so good thinking” (Ain’t that always the case.). It’s generalizing based on a limited number of observations and examples. On the other hand, releasing software and releasing it often is the only way to rid a product of growing pains; launching new software is like taking a previously unused drug to market: you hope it will take care of the problem it’s designed to eliminate without causing unpleasant side effects.

If AD systems are meant to reduce human errors in traffic, they shouldn’t come with their own bag of issues, right? This is precisely what we want to avoid.

The journey to safe automation is a path untraveled. This is all the more reason to advance on it deliberately but methodically. By focusing on a comprehensive approach to safety that balances compliance with standards, understanding human behavior, and capacity for dealing with the unexpected, we’re fighting relentlessly to convince the public of the reliability and benefits of this revolutionary technology. Through patience, transparency, and continuous improvements, we believe that our approach to automation can redefine automotive safety.

We really do.

Stay tuned.

19 April 2023
Credits: Christian Sjögreen, Jonas Ekmark and Fredrik Sandblom


Are you located in China? Follow us on WeChat for regular updates!