A persistent coverage control problem in a complex geometry environment is considered for two-wheeled mobile robots with anisotropic sensor ranges. The persistent coverage control is one of the coverage control methods, which covers the region persistently along the gradient of the time-varying density function. In this paper, we first focus on the persistent coverage control for the two-wheeled mobile robots assuming that the sensing region is anisotropic, and derive a control law which coordinates the rotation and the forward motion. In order to achieve coverage control in the environments with complex geometry, we propose a region assignment planning by employing Monte Carlo tree search for the area allocation. The strength and limitation of the proposed method are discussed based on simulation and experimental results.
In considering the efficiency and/or profitability of a system, a reference value should be set to maximize them. An extremum seeking (ES) control, a type of gradient method, does not require prior information related to the optimization function to find the extreme value of the optimization function. However, this scheme has a problem of taking time to reach an optimal value because the search point is updated repeatedly. This study proposes the database-driven extremum seeking control (DD-ES) method in order to solve the problem and applies the DD-ES to an anti-lock braking system (ABS) model to verify the effectiveness of the actual system case.
Although the number of traffic accidents is decreasing in Toyama prefecture, the number of accidents related to elderly people is more than the average of Japan. Toward prevention of traffic accidents in consideration of coming high-aging society, we propose a way to analyze traffic accidents by using a data analysis method, called formal concept analysis (FCA), which is known to be useful to analyze relationships between data's attributes. We also propose a way to use FCA for a prediction of traffic accidents by using machine learning (ML). It is known that selection of features is important to obtain higher-precision ML models. We use FCA to obtain suitable features for ML.