A simple and efficient method of self-organizing associative data-bases is proposed with a demonstration of application to an autonomous mobile intelligent robot. The system consists of a processing block prescribed and a memory block being blank at the initial state. For a current input signal, the processing block searches for a past input signal that is similar to the current one in the memory block, and recalls the referred information. If the information accompanied with the current input signal does not agree with the recalled one, the new information is added to a data-base in the memory block, with influence reflects on the processing at the next time. Thus, the data-base grows dependently on its states and input signals in a self-organizing way. Since carrier of the data-base organization depends on the series of inputs, each data-base acquires an individual structure. If an input series of sufficient length is given, the relation between input signals and the referred information tends to agree with that of the input series. This system is meaningful in the sense that it gives a conceptual but essential model of intuitive information processing in human's brain and, from a practical viewpoint, in the sense that this system is applicable to a learning mechanism that learns universally any input-output response desired. A camera-type autonomous mobile robot for indoor passages utilizing this learning mechanism is introduced. The brain of robot is constructed on a microcomputer of processor i80286, and behavior of the robot is controlled with a small-size robot language that is supplemented to C-language. A successful experiment result is shown.

View full abstract

In stuties on vision for intelligent robots it is one of the most important subjects to extract from two-dimensional images three-dimensional information such as normals to the object surface and distances from the camera. Numerous methods have been presented so far. A method using coneshaped beams of light has been proposed by the authors. The method can measure the normal vectors and the three-dimensional coordinates of points on the object surface. However, some problems still exist in this method ; for example, uniqueness of the solution, measurement errors, resolution of the system, etc.
In order to solve those problems, the present study presents a new method using both projected patterns and advance of a camera. In this method, the changes of the center positions of the patterns caused from camera advance are used for determining depths from the camera. Then, the determined depths and apparent distortions of the projected patterns are used for the calculation of the local surface normals. Finally, the solution is improved by using the global information of the objects and information of pattern sizes at the central region of the observed scene. Results of the experiements both with simulated images and with real images are also reported. The present method is advantageous over the previous methods in reliability, acccuracy, and resolution.

View full abstract

The hybrid position/force control allows a robot to perform skillful manipulations, while the programming is more complex and intricate, compared with simple position control schemes.
In this paper, an efficient method is developed to eliminate the manual programming and the task interpretation/translation. The operator teaches a given task by “teaching-by-showing”, in which the operator contacts the robot end effector to the environment, and accommodates the contact force. During the operator's motion, the force applied by the operator as well as the position of the end effector are measured. The acquired motion data are then processed and interpreted so that necessary information to generate robot programs is obtained. The choice of control modes as well as reference inputs to the robot controller are derived from the motion data. Then the result is translated into a robot program.
First, the principle of this method is described. The algorithm to interpret motion data is then developed for a simple palletizing job. The method is implemented on a force-controlled direct-drive arm and a personal computer.

View full abstract