This project was designed only for highways with well marked lane lines and daytime driving. While I proved the concept, I never actually let it control the car. It did, however, perform exactly as I expected. The robot is autonomous and based on my SAM algorithm.

An array of 16 light sensors was arranged in a box with a lens such that one sensor "saw" the width of the white lane divider line. The gain was set to distinguish between pavement and the white. The outputs went to circuits that listed the excited cell as a number from 0 to 15. This gave SAM an environment with 16 potential values.

Another sensor at the base of my steering wheel read the wheel position in straight and 3 places left and right. SAM's goal was to steer like me.

Each time a stripe segment excited a sensor SAM would propose one of his seven reactions. That proposal was compared to my own reaction. Identity was perfect and the difference from identity determined the good/bad value of SAMs actions.

Since my own driving is not perfect, SAM learned slowly. It took him typically 18+ miles to drive like me (only more consistently better). After 20 miles I turned SAM from learning mode to "control", though he didn't really control anything.

Apart from SAM's intrinsic functions, for my own information, I logged SAM's performance in "control" against my own driving. SAM did better. This simple and readily duplicated experiment shows how versatile the SAM program is for robotic self-learning.