This study describes a method for generating the online motion of multi-degree of freedom robot. To quickly generate motion, we adopt an approach that accumulates a robot's past motion experiences and subsequently selects and modifies the appropriate motions from this experience. The key aspect of our method involves using swept volume. As a pre-processing step, the robot's movements for reaching various target postures in a range of environments are derived using a conventional motion planning method. At this time, the volume occupied by the robot, i.e. the swept volume, is recorded using the destination posture and generated motion. Then, using a new target hand pose, the system searches past experience data closely linked to the target posture and selects the motion with a swept volume that does not interfere with the current environment. With our implementation of swept volume, collision detection is achieved only one bitwise operation. Moreover, it is possible to generate a larger range of behaviors outside the collected behavior data by connecting multiple motions and swept volumes. Our method's effectiveness was then confirmed by motion generation experiments using a mobile manipulator.
展开▼
