In this paper, we propose a smart teaching system for many complex tasks achieved by a seven degrees-of-freedom robotic manipulator. In this system, we basically use a sliced 2-D configuration plane because of no occlusion. In general, an enormous number of such planes exist in seven dimensional configuration space. To select some fruitful planes from all the planes, we develop a human support system based on efficient path finding and obstacle evaluating algorithms in each 2-D plane. The efficiency of this human computer cooperation system is compared with that of the model-based path-planning algorithm A*. Selecting a near-optimal path in our system is absolutely faster than calculating its optimal path by A*. This means that a human ability understanding a task is quite powerful especially for a seven-degrees-of-freedom robotic manipulator.

In this paper, we propose three types of sensor-based path-planning algorithms and compare them as the image-based path-planning algorithm for a huge search space. In the general image-based path-planning, a camera mounted on a tip of a manipulator is seeking for an objective image while the manipulator is controlled in a joint space. As long as the number of degrees-of-freedom of a manipulator increases, the search space becomes exponentially huge. Though the sensor-based path-planning algorithm is a noncombination search, it sometimes spends much time to escape from a valley minimized by a local minimum. To overcome this problem, we use a modified version of the randomized algorithm as the best image-based path-planning algorithm for a huge search space. This tendency is checked by several simulation results. In addition, the version can be easily applied for a real problem as the classic visual servoing (the steepest descendent method) with memorizing a set of visited points and another set of their neighbor points, as well as generating a sequence of random motions.