Robot programming by human demonstration: the use of human inconsistency in improving 3D robot trajectories

Programming by human demonstration is an intuitive method of robot programming, in which the programmer demonstrates how a task is performed using a teaching gripper that measures human motion, and the data gathered is used to generate the robot program. A direct duplication of the demonstrated trajectory would result in unsatisfactory robot motion due to human "wiggles" and unnecessary motion. Here, human inconsistency is not treated as noise that should be eliminated from the analysis, but is used to gain information regarding the task accuracy requirements. A task of moving a part to a desired location while avoiding stationary obstacles is demonstrated. Variation between the demonstrations is interpreted as human inconsistency and defines a range of acceptable obstacle free robot motion. In previous work, a method was presented to generate robot trajectories for two dimensional motion. Here an alternate algorithm is presented that is extended to include three dimensional motion. The presence of human inconsistency is used to improve robot performance by generating robot trajectories that are shorter than the demonstrated ones and with the addition of a buffer between the robot and the obstacles.>