ADAPTIVE LEARNING OF HUMAN MOTION BY A TELEROBOT USING A NEURAL-NETWORK MODEL AS A TEACHER

This paper describes a model and experiment in which human motion data is used to teach a laboratory telerobot simple human skills. The prototype learns an adapted motor skill involving pick and place tasks in a microblock world environment. We choose place and pick tasks because of their repetitive and dynamic motor motions involved during both forward and backward task execution. By assuming the skill transfer takes place implicitly, we use human skill data to train the robot via a neural network simulation model. The method has an advantage over the classical state-space control models in that the model is adaptive to changes in human input data.