An interesting article from Robotics Trends on the problems which will keep self-driving cars off the road for many years, despite the hype that their arrival is “just around the corner”
“Crappy US Roads Major Roadblock for Self-Driving Cars?”
Poor lane markings and uneven signage on the three million miles of paved roads in the US are forcing automakers to develop more sophisticated sensors and maps.
In other words, the reason we don’t have robots is that the world isn’t up to snuff. It’s not that robots just aren’t smart enough to navigate the real world. Fix the world (by making everyone paint high-resolution lane markers) and robots can continue their expected takeover.
Now, imagine if we applied this to wheeled robots. Humanoid robots aren’t practical at present, as demonstrated by recent DARPA challenges. They blow up in complex environments with poorly marked walk regions and stairs. Is the solution to rip out all building stairs and clearly paint lines on the floor the robots can follow? Or, should people all move around in wheelchairs, so they move in the same way robots do? Does this fix the problem?
What this kind of reasoning tells us (which the article isn’t guilty of, btw) is that a passion for having robots around requires we don’t make smart robots. Rather, we re-engineer reality so that robots can understand it. This will allow “machine intelligence”.
In fact, changing highways everywhere would be so massively expensive and time-consuming that it won’t happen. Period. Self-driving cars will be practical only when they can navigate current roads.
The real story was relayed to Congress via Mary Cummings, the head of Duke University’s robotics program. A good podcast here:
In other words, the real headline should be
“Crappy Robots Can’t Navigate Ordinary US Roads”