This paper describes a new preview lateral control algorithm for autonomous vehicles with machine vision and experiments with an AGV conducted to demonstrate the feasibility of the algorithm. Compensation of the delay due to the time for image processing is also considered. The steering control algorithm uses a reference line along a path for lateral control. It finds the control by approximating the reference line with a cubic curve, a coefficient of which gives the steering control. The algorithm uses the whole information in the two-dimensional field of view, which enables robust and smooth lateral control even if a reference line is not aligned or is composed of only straight segments.
A stabilization problem is considered for linear time-invariant multi-input multi-output systems with variable operating conditions. The plant is assumed to be modeled as an interpolation of two nominal plants described by proper stable factorizations of transfer function matrices. To stabilize the closed-loop system, two kinds of controllers are proposed. One is a fixed controller, and the other is an interpolation of stabilizing controllers for the nominal plants. For each case, sufficient conditions for stabilization are reduced to solvability of certain H∞ control problems. Design algorithms for the proposed controllers are presented in the state space. The relation to the simultaneuos stabilization problem for two nominal plants is also considered.
The collocated feedback is the direct negative definite feedback for flexible structures of velocities and/or displacements measured by sensors collocated with actuators, and preserves robust stability of the closed-loop systems. But, generally it does not insure sufficient low sensitivities for uncertainties or input disturbances ; therefore, an improvement for low sensitivities is necessary. In this paper, we show the condition of the Robust Model Matching to increase the dampings of the elastic oscillation by the collocated feedback and lessen the sensitivities simultaneously. Then we show experimental results using a flexible arm apparatus and discuss the performance.
A new identification system of large-spatial air pollution patterns using a neural network and a source-receptor matrix, is described. The neural network used in this paper can identify a nonlinear system whose structure is very complex. In the previous identification system of a large-spatial air pollution patterns, a GMDH algorithm and a source-receptor matrix are used. The prediction results obtained by using the new identification system are compared with the results obtained by using the previous identification system. It is shown that the new identification system in this paper gives better prediction accuracy as compared with the previous identification system.
In this paper, fractal basin boundaries are investigated in connection with a class of one-dimensional nonlinear discrete-time systems. Noting that the basin is an invariant set of the nonlinear function, describing the system dynamics, conditions, under which the Hausdorff dimension of boundaries of invariant sets is positive, are obtained and the existence of fractal basin boundaries is shown. Furthermore, it is demonstrated that if periodic points with period three exist, then fractal basin boundaries appear. Secondly, the result obtained is applied to explore, the basin boundary of a class of one-dimensional nonlinear sampled-data control systems. Illustrative examples together with numerical experiments show the existence of fractal basin boundaries. These theoretical and numerical results reveal that in order to determine the sampling period, it is necessary to take into account the structure of the basin of the equilibrium.