As most of the car industry have scrambled to develop fully driverless car technology over the last few years, Toyota has taken a pointedly contrarian tack.
“None of us in the automobile or IT industries [is] close” to solving the challenge of fully autonomous vehicles, said Gill Pratt, CEO of the Toyota Research Institute, in a speech at the Consumer Electronics Show on Monday.
As a result, Toyota has taken a two-track approach to its own self-driving vehicle research. On one track, dubbed “Chauffeur,” Toyota is working to develop fully autonomous vehicles similar to those being created by Alphabet’s Waymo, GM’s Cruise, and other companies.
On the other track, dubbed “Guardian,” Toyota is working on more sophisticated driver-assistance products—products that can intervene when a driver makes a potentially deadly mistake. If a child darts out in front of a vehicle, for example, the car might automatically swerve to avoid hitting her.
At this year’s CES show in Las Vegas, Pratt touted the company’s progress in developing Guardian technology. Guardian is still far from being a shipping product. But at Monday’s presentation, Pratt disclosed plans to license it to competitors—an approach the car maker dubs “Guardian for all.” Toyota is still figuring out what this licensing program might look like—and whether the company will license hardware, software, or some combination of the two.
The challenge of Guardian technology
Up to a point, the Guardian approach makes perfect sense. Indeed, you can view automatic emergency braking technology, which is rapidly becoming an industry standard, as an early example of Guardian technology.
And there’s no doubt that today’s emergency braking systems have room for improvement. Last year, for example, several Tesla vehicles slammed into stationary objects while Autopilot was enabled. One of these crashes claimed the life of its driver.
Today’s driver-assistance systems are simply not designed to prevent all accidents. Many emergency braking systems are designed to ignore stationary objects when the car is traveling at freeway speeds. So if Toyota can develop more sophisticated emergency braking systems, that could indeed be a valuable contribution.
How much further the Guardian approach can go is not clear. It’s one thing to slam on the brakes in the face of an imminent collision, but it’s another thing for a vehicle to override its driver to take complex evasive maneuvers.
In one 2017 video, for example, Toyota demonstrates a prototype Guardian vehicle automatically changing lanes on a freeway (actually a private test track) to avoid hitting a wheelbarrow that’s been left in the middle of the road. It’s fairly easy to program a car to perform this maneuver on a closed track. But doing this kind of thing on a real freeway is fraught with danger.
Before changing lanes, the vehicle needs to be very, very sure it understands where the lanes are and what nearby vehicles are likely to do. If the software misunderstands the situation, it could cause an accident rather than preventing one.
An unexpected lane change could also disorient the car’s driver and indirectly cause an accident. Imagine, for example, if you glance down at your smartphone just as your car initiates a lane change. When you look back up at the road, you might not immediately recognize the car’s position. This could lead you to make a mistake.
A sophisticated driver-assistance system may be able to overcome all these challenges. Good-enough sensors and software may be able to fully understand the car’s environment and accurately predict what other road users will do. Sophisticated driver-monitoring systems and a good user interface could ensure the driver is never caught off-guard by the vehicle’s interventions.
But the larger question is whether this is really an easier problem to solve than building a fully self-driving car. If Toyota develops software that can confidently make lane changes in a variety of tricky emergency situations, will it really be that much harder to just have the software drive the entire route?
Toyota will probably discover some scenarios where a car can intervene with minimal danger of causing a crash or confusing the driver. For example, Pratt showed an example in which a car suddenly emerges a few feet onto the roadway from between parked cars:
It’s not crazy to think that, in this scenario, the Guardian system would be able to quickly verify that there are no obstacles to the vehicle’s left and then swerve a few feet in that direction—just enough to avoid a collision. Guardian could then return to the original travel lane a few seconds later. That could prevent some crashes with minimal risk.
Still, this is far from a trivial problem. It may turn out that a system sophisticated enough to reliably identify these situations and handle them safely isn’t that much easier to build than a system that just drives the entire route itself.