Google has been working on robots that learn to walk autonomously; they teach themselves through a kind of trial and error.
First, they bounded the terrain that the robot was allowed to explore and had it train on multiple maneuvers at a time. If the robot reached the edge of the bounding box while learning how to walk forward, it would reverse direction and start learning how to walk backward instead.
Second, the researchers also constrained the robotís trial movements, making it cautious enough to minimize damage from repeated falling. During times when the robot inevitably fell anyway, they added another hard-coded algorithm to help it stand back up.
Through these various tweaks, the robot learned how to walk autonomously across several different surfaces, including flat ground, a memory foam mattress, and a doormat with crevices. The work shows the potential for future applications that may require robots to navigate through rough and unknown terrain without the presence of a human.
I was strongly reminded of the learning robots from Callahan and the Wheelies, a 1960 short story by Stephen Barr.
"...they're motivated first by a random device and then they learn. The lines of connection in the graphite-gel that turn out the most successful remain like a printed circuit and then if occasion arises, they overprint them. My whole idea is to get away from a machine with a set of prearranged instructions, and let them teach themselves by trial and error."