TRAJECTORY GENERATION FOR SENSOR-DRIVEN AND TIME-VARYING TASKS

In on-line robot trajectory generation, a connecting polynomial is nor mally used to remove discontinuities in velocity and acceleration betw een adjacent path segments. This article presents a new technique for performing such transitions in which adjacent path segments are ''blen ded'' together with excess acceleration being removed using an estimat e of the initial path velocities. Because this method requires no adva nce knowledge of the path segments, it can handle situations where the paths are changing with time (as when tracking sensor or control inpu ts). The method can also be used to adjust the spatial shape of the tr ansition curve (such as to have it pass around or through the ''via po int''), which may be necessary to handle constraints imposed by differ ent types of manipulator tasks. When the blended paths are nonlinear i t is possible to set a tight bound on the resulting transition acceler ation. The blend technique works directly for vector trajectories and can be modified to handle 3-D rotational trajectories. A simple trajec tory generation algorithm is presented as an illustration.