From Autonomous Vehicles That Crash To Autonomous Vehicles That Learn

Bleeding Edge

from the lessons-learned... dept

While some people considered DARPA’s Grand Challenge race a flop when the autonomous vehicles failed to get very far, many involved thought that it was a good start, and with a little more time in the shop they could build autonomous vehicles that could successfully navigate the course. DARPA, however, seems to want to go in another direction. It seems like a bit of a kneejerk reaction away from where things were already headed, but now they say they want to focus on autonomous learning vehicles. The idea is that a smarter autonomous vehicle would learn to recognize a pothole and avoid it the next time one came around. All this really does, though, is add to the complexity of the vehicle, without necessarily leading to much benefit. Learning systems have a mixed history with most systems not doing all that well. To achieve this plan, though, DARPA plans to hold a series of races between a “learning” vehicle and a regular autonomous robotic vehicle, with the goal being for the learning vehicle to get much better each time. It’s a little unclear if this new direction is separate from the regular autonomous Grand Challenge or an attempt to replace it. Still, some of those involved with the Grand Challenge say their systems will contain some amount of adaptive learning. However, the general consensus is that, for right now, it’s more productive to focus on autonomous machines and let the humans do the learning.