Task Oriented Machine-Learning and Review

We propose an optimization algorithm to execute a previously unlearned task-oriented command in an intelligent machine. We show that a well-defined, physically bounded, task-oriented problem can be solved if a machine has knowledge of the goal of the task, is capable of determining the advantage of a decision, has knowledge of itself, has knowledge of the perimeters of the problem space, and is provided the basic physical skills to execute the task. The algorithm proposed operates on a cycle composed of learning, reviewing, and optimization. We use the classic robot path finding problem in an initially unknown environment to illustrate the algorithm.