One of the remaining challenges for autonomous cars is figuring out a way to handle that long tail of weird edge cases that can randomly happen in the real world. Another challenge that remains is figuring out how to handle that huge population of weird beings called humans, who behave pseudorandomly in the real world.
We've seen potential solutions to both of these problems. The first, meant to cover the long tail of situations that are outside the experience and confidence of an autonomous system, can involve a remote human temporarily taking over control of the vehicle. And with the right hardware, that human can solve the challenge of interacting with other humans at the same time.
Vehicle-human interaction is a tricky thing. Modeling humans works to some extent, and autonomous vehicles can predict with reasonable accuracy the path that a human on a sidewalk, for example, is likely to take in the immediate future. Where things get complicated is when humans interact directly with drivers: The classic example is a crosswalk, where a pedestrian will make eye contact with the driver of a vehicle to make sure that they're seen, and that it's safe to cross.
Without a person in the driver's seat, there needs to be some other way to communicate, which is why Drive.ai, for example, mounted LED displays on its self-driving vehicle that could display messages in text and icons. But a truly robust solution needs to be much more human-like, and Ilhan Bae, a researcher and futurist at the Korea Advanced Institute of Science and Technology (KAIST), thinks that a combination of telepresence and vehicle autonomy is the way to make that happen.
This is a little weird, granted, but it's a potentially effective way of addressing both the unlikely events problem and the human interaction problem. A 5G connection makes high definition streaming possible between the 360-degree camera on the roof of the vehicle and the remote user. And the projector inside the vehicle is bright enough for the image to be visible even in direct sunlight. The KAIST researchers say that this sort of system is uniquely valuable for Asian countries, whose driving environments are often crowded and chaotic—scenarios very different from the situations that autonomous cars are comfortable driving in right now. So, pedestrian-to-driver and driver-to-driver communication plays a much more significant role than elsewhere.
Beyond just communication, the KAIST researchers say that these telepresence “persona cars” provide additional value as adaptable extensions of remote users. According to Bae, “When a police officer is connected, it becomes a police car, when an insurance investigator connected, it becomes an insurance company car, when reporter is connected, it becomes news reporting car.”
The telepresence car is subject to certain compromises—on one hand, you lose out on some of the benefits of full autonomy (operating without a human driver), and on the other, the level of mobility offered by a car is, by definition, restricted to areas where cars can travel. Realistically, this system may have the most utility as a way for humans to effectively take control of otherwise fully autonomous cars when necessary, making them flexible enough to operate in places where autonomy struggles.
Evan Ackerman is a senior editor at IEEE Spectrum. Since 2007, he has written over 6,000 articles on robotics and technology. He has a degree in Martian geology and is excellent at playing bagpipes.