The Autonomous Car: How Machine Learning Broke Through Limits That Engineering Couldn’t Get Past
Introduction
The autonomous car began as an opportunity that required breaking all kinds of limits: engineering, navigation, adjusting to traffic conditions, distinguishing objects, predicting what those objects might do, reacting in time, calculating quickly and juggling a vast number of ever-changing variables. The developers used more and more computer power to address these needs. But the initial bounding limit turned out to be very fundamental; rule-based computers don’t have pattern power.
From the beginning systems with pre-programmed data definitions and structured procedures – rules – dominated the use of computers for everything from standard business data processing applications to even AI. And so it was logical that the early approach for autonomous cars was to define rules, derived from experts, and embed those rules in the software in the form “If..Then… Else…” “If the car is stationary and a pedestrian crosses ahead, remain stationary. Otherwise, check the brake is off ..If it isn’t, then…” But that did not work, as the first Grand Challenge competition for autonomous cars made very clear.
- Keyword(s):
- innovation 176
- technology 76
https://kenaninstitute.unc.edu/publication/the-autonomous-car-how-machine-learning-broke-through-limits-that-engineering-couldnt-get-past/
