Robots baffled by optical illusions

From Crave: The gadget blog:

In theory, robots aren't designed to make mistakes. But a University College London (UCL) project team is hoping errors in how software "sees" optical illusions can make robots more like humans--mistakes and all. Project leaders Dr. R. Beau Lotto and David Corney at the UCL institute of Ophthalmology say the study provides unprecedented insight into how the human eye can be fooled by lighting and shading. Instead of simulating the human brain, the software simulates learning patterns from past visual experiences. The UCL Institute of Ophthalmology study recreated the vision errors using software that "learns" colors and shading based on thousands of images. After the software was trained to predict the shade of colors from those images, it was subjected to shade-based optical illusions.

Read the whole article