Mazda May Have Cracked One Of Autonomy's Biggest Near-Term Challenges

I'd like to confirm all of this in much more detail, but this ability is a crucial part of getting past L2

We may earn a commission from links on this page.
Image for article titled Mazda May Have Cracked One Of Autonomy's Biggest Near-Term Challenges
Image: Mazda

As long as semi-automated cars require a driver to be ready to take control at a moment’s notice, we’ll be stuck at Level 2 autonomy. The only path past this requires having cars that are able to, if they find they need human input and are unable to get it, disengage safely and get out of everyone else’s way, on their own. So far, no automaker has demonstrated this ability, but a new concept from Mazda claims to, though the ability is framed in terms of dealing with medical emergencies.

Currently, the best semi-automated Level 2 vehicles do is to slow down to a stop in their lane of traffic and turn on the hazard lights. There seems to be a lot of confusion about this behavior, with many people believing the car can get safely off the road, but so far this hasn’t proven to be the case, even for Tesla, which I tested myself.

Level 2 semi-automated systems all need some form of driver monitoring because these systems can disengage without warning, requiring the driver to take over. The problem is none of these driver-monitoring systems are foolproof, and abuse of these systems is all too common.


Part of the problem is inherent in Level 2 systems conceptually, because humans simply aren’t good at “vigilance tasks” like paying full attention to a car that’s doing 80 percent or more of the driving task.

So, the solution is that the car needs to be able to get out of an active traffic lane and somewhere safe when it detects a non-responsive driver, or needs to disengage for other reasons. Mazda’s Co-Pilot Concept car seems to be able to do just that, at least based on what they claim, and what’s shown in this video:

In that video, the car appeared to continue to drive until it found that convenient turn-out lane, which it seems it has been programmed to identify.


A bit more information was told to the Seattle Times by Mazda:

Mazda told reporters recently it has been working with medical experts, including Tsukuba University Hospital, researching the collected image data to figure out what a healthy driver looks like, as opposed to an incapacitated driver, suddenly slumped forward over the steering wheel.

Once recognizing a problem, Co-Pilot Concept, which has yet to have an official name, will bring that car to a stop in a safe spot, such as the curb of the road, as quickly as possible.

The car will be honking, with blinker and hazard lights flashing, according to Mazda, although the exact warning signals are still undecided. An emergency call to the ambulance and police will also get relayed.


Here, the claim is that the system will “bring that car to a stop in a safe spot, such as the curb of the road,” which sounds like they mean, at the very least, the car will be out of an active traffic lane if possible.

I reached out to Mazda for further clarification: can it change lanes on a highway to get to the shoulder to stop? What will it do if no suitable shoulder is available? I have yet to hear back from Mazda with any details, but I’ll update as soon as I hear something.


Japan’s aging population seems to be the main motivation from Mazda to develop this system, which will also be able to identify head motions and driving behaviors that may indicate a loss of consciousness or other medical problem is imminent.

Being able to safely deal with an incapacitated driver is a huge deal, and has, as I’ve mentioned, implications far beyond just dealing with medical issues. If the system works, there’s really no reason why other automakers couldn’t implement similar safeguards into their semi-automated systems, helping to push them past Level 2.


Dealing with an unresponsive driver is only part of the equation to get past Level 2, of course; being able to safely pull off an active traffic lane when the car itself determines its own systems have a problem or issue is the harder task to solve, because if the AI driving system or hardware is impaired in some way, how can it safely get off the road without demanding immediate human intervention?

Figuring out how to pull that off may require robust car-to-car communication or infrastructural help, but at least it looks like some progress is happening.


I’m curious to see more of how Mazda’s system works. I hope they get back to me.