Darrin C Bentivegna, Christopher G Atkeson, Gordon Cheng
Learning tasks from observation and practice
This paper presents a framework that gives robots the ability to initially learn a task behavior from observing others. The framework includes a method for the robots to increase performance while operating in the task environment. We demonstrate this approach applied to air hockey and the marble maze task. Our robots initially learn to perform the tasks using learning from observation, and then increase their performance through practice.