The word “groundbreaking” is frequently used when describing autonomous vehicle technology, and nobody doubts that driverless cars should change the world of transportation as we know it.
But there’s another word that Bloomberg has recently applied to the friendly self-driving car and that’s “annoying” … kind of like, we’re grateful the technology exists, but really don’t want to get stuck behind one.
That’s because right now, most autonomous vehicle systems lack what one might refer to as “practical common sense” in many situations. Bloomberg describes test vehicles as being “maddeningly slow” and prone to paralysis when encountering normal transit variables like joggers or crosswalks.
Uber’s self-driving car program was working on technology to minimize excessive braking around pedestrians, and it was in use when an autonomous Volvo fatally struck a woman walking her bike across a dark street. So, clearly there’s a lot of R&D left to be done here, but in the meantime, a startup called Perceptive Automata believes artificial intelligence can be used to teach autonomous vehicles to “think” more intuitively, or, as a human would.
The company is said to be attempting to teach the vehicles how to predict human behavior by trying to model how humans do it. Image recognition tests – the same ones used in psych experiments – are being used in training machines in the same fashion, loosely, as the brain works.
Real people view video and images and determine what they believe the pedestrians are about to do. For example, they add intuition to identify the difference between a bicyclist on the corner about to ride across a path, versus a bicyclist on the corner stopped and looking at his phone. This all gets fed into a “reference library” to better help the machines understand what’s actually happening.
The system is even being designed to account for regional differences, so hopefully the downtown Chicago iteration will know to lay on the horn the second the stoplight turns green.
Perceptive Automata recently raised another $16 million in funding, and is said to be working with automakers and suppliers across the world – including in the U.S. The company believes it may be able to integrate the system into production cars by 2021.