ABSTRACT
This chapter addresses the engineering problem of creating robotic agents that perform tasks in an environment. One motive for making artificial agents adaptable is that many natural agents are adaptable, and there is a strong analogical similarity between natural and artificial agents. The chapter focuses on the long-term adaptation of behavior to suit general properties of the embedding environment. One way to view the problem of constructing adaptable behaviors for agents is as a reinforcement learning problem. In reinforcement learning, the goal of the agent's designer is for the agent to learn what actions it should perform in which situations in order to maximize an external measure of success. The chapter concludes that there is a spectrum of situations of information gain, ranging from what is commonly described as perception to what is commonly described as learning or adaptivity.
