Servo - May 2015

Like a child randomly reaching for a goal (like a cookie jar), a lot of things we learn just by being in the environment. You have to get your robot “out there” before it can begin to have its own experiences from which to learn. Isn’t this the way you learned things?

Random and hap-hazard at first.

Ultimately, you have some memory of what to do (and what not to do).

Finally, you generalize this information and use it in other situations or environments. This is how we learn. Machines can learn the same way ... through experience. The Arduino-based Junior’s primary mode of expression is movement, but not just any motion pattern — he has no goal other than to move forward or backwards. Again, these are the only two motion patterns with significant magnitude. To that end, he will try everything (all nine motion patterns) in order to reach his goal. In How to Program Your Own Self Programming Robot (Figure 2), David L. Heiserman describes three levels of machine intelligence: Alpha, Beta, and Gamma.

Alpha can be described as random motion. The robot will randomly try motion patterns or codes until he stumbles upon 4 (forward) or 8 (reverse). Then, when he stalls or bumps into something, he will again try random motion patterns until he re-discovers forward or reverse.

Beta level intelligence remembers the successful response for a

When we last left off in the February issue of

SERVO Magazine,

"Rodney Junior" was happily bouncing around in his playpen, randomly exploring his world and eventually learning the two motion patterns that maximize distance traveled within an environment; namely going forward and backwards. Recall that there are nine possible motion codes in a differential drive system including stop (Figure