 Directed robots run by themselves but can be trained. Like dogs, they learn the meaning of commands by getting rewards. In the Directed Robots Lab at Sandia National Laboratories, a surveyor SRV-1 robots named Rolf is learning to hide and seek. Hide means to find a picture that is either all light or all dark. For him, seek means to find a picture with lots of light and dark. When he first starts up, Rolf doesn't know how moving will change what he sees. He just tries different movements and remembers what happens each time. Eventually, he learns that moving forward toward the wall is a good way to hide and that backing up is usually a good way to seek. Sometimes, Rolf gets bored or curious and tries something different. This helps him figure out the best way to do what he is trying to do. Most robots know a lot about themselves, like the size of the room they're in and whether they have wheels or arms. Rolf doesn't know any of that. He doesn't even know that his eye is an eye and not a nose. He doesn't know how far he'll turn when he spins. He only remembers what his sensors tell him and what rewards he gets for hiding and seeking. Rolf's brain is a computer program called S-Learning. S-Learning doesn't care what kind of robot it is connected to. It can be a brain for lots of different kinds of robots. S-Learning learns whatever it gets rewarded for. Like a baby, S-Learning uses all its past experiences whenever it has to learn to do something new. It might even be able to help a robot do hard things like run on two legs or learn to talk. But for now, Rolf is still learning how to fetch.