To evolve robot navigators that generalize well, we should evaluate the navigators using as many environment patterns (EPs) as possible during evolution. On the other hand, to reduce the computational cost, we should consider as less EPs as possible. A good compromise is to use some well selected EPs. However, selection of good EPs is in general a very hard problem. To solve this problem, we have proposed an improved co-evolutionary algorithm (ICEA) for evolving the navigators and the EPs together. So far, we have verified the ICEA with neural network (NN) navigators. The NN navigators, however, are black-boxes that cannot be understood easily. In this paper, we try to evolve fuzzy rule (FR) navigators with the ICEA. Simulation results show that the FR navigators evolved by using ICEA generalize as well as the NN navigators, and they are much more comprehensible.

To evolve robot navigators that generalize well, we should evaluate the navigators using as many environment patterns as possible during evolution. To reduce the computational cost, however, we should use as few environment patterns as possible. It is difficult to know in advance what patterns can evolve good navigators. To solve this problem, we study two different approaches. One is the co-evolutionary algorithm (CEA) that evolves the navigators and the environment patterns together, and the other is to select good environment patterns through incremental evolution (IE). Detailed considerations in using CEA and IE are described in this paper, and their efficiency is compared with each other through simulations.

To evolve robot controllers that generalize well, we should evaluate the controllers using as many environment patterns or test cases as possible. On the other hand, to evolve the controllers faster, we should use as few environment patterns as possible in evaluation. It is difficult to know in advance what environment patterns can produce good controllers. To solve this problem, this paper studies co-evolution of the robot controllers and the environment patterns. To improve the effectiveness of co-evolution, we introduce fitness sharing in the population of environment patterns, and the inter-generation fitness in selecting the robot controllers. Simulation results show that the improved method can get much better robot controllers than standard co-evolutionary algorithm.

A general-purpose two-dimensional process simulator OPUS (Oki Process simulator for ULSI Structures) for arbitrary structures is discussed. By separating a preprocessor from a main processor, OPUS can simulate any kind of impurity or material with up to eight impurities and ten materials. OPUS utilizes a finite-difference method for diffusion analysis without any coordinate transformation. In order to minimize errors due to discretization around the curved boundaries, OPUS has introduced two kinds of boundaries. The first is true boundaries by strings. The other is quasi-boundaries formed through simplifying but retaining the shape of true boundaries as long as possible. The quasi-boundaries and resulting dummy cells are used for discretization of the diffusion equation. General features of OPUS are first introduced with special emphasis on dummy cells. Errors caused by dummy cells are discussed. The capacity of OPUS is then demonstrated through application to a trench process and to a full MOS process with a sidewall.>