A major goal in robotics has been to give robots the capacity to learn and reason about new tasks without explicitly programming this new information. This approach takes the programming of robots away from writing tedious code, and transforms it into a process of learning through experience. Experience-based learning is employed by human beings, primates, mammals, and a large number of other animals. In particular, there is much evidence for the existence of an adaptive working memory in primates. Such memory is closely tied to the learning and execution of tasks, as it contributes to decision-making capabilities by focusing on essential task information and discarding distractions. The employment of such a biologically-based adaptive working memory within a robot has not been previously investigated. Our goal is to develop an adaptive working memory to support the following: Focus attention on the most relevant features of the current task. Support generalization of tasks without explicit programming. Guide perceptual processes by limiting the perceptual search space. Provide a focused short-term memory to prevent the robot from being confused by occlusions. Provide robust operation in the presence of distracting events. We propose to integrate the adaptive working memory structure into a robot in order to provide a solid context for exploring the issues of task learning. The adaptive working memory will be tested using a series of evaluation tasks on both humanoid and mobile robot systems.
