Semantic learning by an autonomous mobile robot

This paper describes the design and implementation of a learning system for control of an autonomous mobile robot. The robot learns reactive behaviors that allow it to retreat from potential collisions and to explore its environment by seeking out nearby objects. No external teaching input is required. Results from experiments with a real robot are presented. The learned reactive behaviors become the basis for the acquisition of more complex behaviors. Sensory/motor states are classified and then associated with lexical items to form a simple command language which is then used to direct the robot.