Yoshiaki Ichikawa

Neural network application for direct feedback controllers

The author presents a learning algorithm and capabilities of perceptron-like neural networks whose outputs and inputs are directly connected to plants just like ordinary feedback controllers. This simple configuration includes the difficulty of teaching the network. In addition, it is preferable to let the network learn so that a global and arbitrary evaluation of the total responses of the plant will be optimized eventually. In order to satisfy these needs, genetic algorithms are modified to accommodate the network learning procedure. This procedure is a kind of simulated evolution process in which a group of networks gradually improves as a whole, by crossing over connection weights among them, or by mutational changes of the weights, according to fitness values assigned to each network by a global evaluation. Simulations demonstrate that these networks can be optimized in terms of various evaluations, and they can discover schemes by themselves, such as state estimation and nonlinear control.

A hybrid locomotion vehicle for nuclear power plants

A hybrid locomotion vehicle for use in remote maintenance of a nuclear power plant is proposed. The vehicle has five locomotion devices, each consisting of a supporting leg and a driving wheel. Each device has three degrees of freedom: leg lifting, wheel revolution, and wheel steering. These degrees of freedom and the inclusion of touch sensors for obstacle detection enable flexible locomotion such as stepping over rises and going up and down stairs in the plant. Two slope sensors set at orthogonal directions maintain vehicle posture through vertical leg motion. The control system, based on a microcomputer, supervises all of these motions. Laboratory tests with a prototype vehicle demonstrate successful locomotion.

Retaining diversity of genetic algorithms for multivariable optimization and neural network learning

Methods to retain diversity of the allele distribution in the search for genetic algorithms (GAs) are presented. The authors seek a technique to prevent premature convergence and refine the performance of GA for use in multivariable optimization and unsupervised learning of neural networks. An integer string representation for chromosomes is defined which is well fitted to this usage. The diversity of each locus and rareness of a chromosome are evaluated based on the distribution of alleles in a population. The fitness of a chromosome is adjusted with the rareness so that rare chromosomes will be likely to survive. Mutation width is introduced to adjust the effect of mutation which can generate rare chromosomes. By dynamically changing mutation width at each locus according to the diversity, prematurity can be avoided while conserving effective convergence. Case studies with problems of neural network pattern matching and unsupervised learning of a neural network which controls an inverted pendulum are discussed.<<ETX>>

A heuristic planner and an executive for mobile robot control

A path planner and an execution system are proposed for automatic indoor guidance. The planner creates a route plan in two levels of hierarchy. The first level selects doors to pass through, minimizing the total number of doors, and the other generates a near shortest trajectory inside a room. In order to reduce search space and computation time for the planning, heuristic pruning strategies are used in a path search procedure. The execution system performs hierarchical plan repair with regard to the plan levels against unexpected events by detecting obstacles. Simulation tests are conducted to evaluate its execution behaviour with various representations of terrain, unknown obstacles, and an omnidirectional vehicle robot equipped with a sonic range finder for obstacle detection. The results exhibit the effectiveness of simple heuristics in combination with a sonic range finder that provides an adequate amount of data from multiple viewpoints.

A parameter tuning method using genetic algorithms for an elevator group control system

This paper describes a new parameter tuning method for an elevator group control system. In an elevator group control system, tuning parameters describing elevator use and the building environment are important. This new tuning method uses improved genetic algorithms to follow environment changes and to give robustness. Keeping diversity is important in following environment changes. The paper also describes a method of adding random chromosomes and a method of adding other evaluations chromosomes to keep diversity. Their effects are shown by simulations.

A parameter tuning method for an elevator group control system using a genetic algorithm

This paper describes a new parameter tuning method for an elevator group control system. In such a system, tuning parameters describing elevator use and the building environment are important. The new tuning method uses improved genetic algorithms to follow environment changes and to give robustness. Keeping diversity is important in following environment changes. The paper also describes the addition of the random chromosome method and addition of other estimation methods to maintain diversity. Their effects are shown by simulations.

One-dimensional self-reproducing robot

A design concept for one-dimensional self-reproducing robots is presented. A self-reproducing machine is composed of separable parts (cells) and possesses genetic data which contains information to decide the structure of the machine body. The design concept has the following characteristics: (1) the cells specialize themselves according to their position in the body; (2) the machine grows in size by capturing isolated cells, connecting them to its body, and transferring the genetic data to the captured cells: and (3) the machine splits apart in the middle and makes two independent halves after growing to twice its initial size. Experiments have demonstrated the self-reproduction of mobile robots, each of which is composed of multiple cells connected to one another in a row, one-dimensionally.<<ETX>>

Joint connection mechanism for reconfigurable manipulator

A reconfigurable dual-arm manipulator than can automatically change the configuration of one arm has been developed for use in maintenance and repair tasks for nuclear plant facilities is described. The development of hardware and force control for the reconfigurable arm is discussed. A design in which controller parameters can be chosen easily with no need for detailed kinematic analysis is proposed. Experimental results are given.<<ETX>>

Intelligent manipulator for remote maintenance and inspection

An intelligent dual-arm manipulator has been developed which can autonomously adapt to various tasks and environments in future plants. The manipulator makes task plans by searching for a collision-free path to the manipulated objects, and can optimize its arm structure by changing joints according to the plans. The typical planning time for an optimal collision-free path, utilizing a heuristic graph-search algorithm, was about one-tenth of that by a conventional method. The manipulator could autonomously reconstruct its left arm joints in five minutes by using its right arm.<<ETX>>