Recently, transportation of supplies using Unmanned Aerial Vehicles (UAVs) such as quadrotor drones has been attracting attention. However, due to limited battery capacity and payload, it is difficult to transport supplies over long distance in a single flight. Therefore, this paper proposes a route planning method in which the battery is recharged or replaced via charging stations along the way, depending on the transportation route. In addition, along with the distance between stations, factors such as the wind direction and the number of stops at the charging station are considered in the calculation of the cost of transportation. Multi-objective Dijkstra algorithm is proposed to select a route that minimizes transportation costs from the origin to the destination and back. The proposed method was applied to a transportation environment with wind and the results show that the algorithm was able to optimize the delivery route with criteria of distance, wind effect and the number of stops.

This paper proposes a model compression method of reducing the number of nonlinear activation functions of continuous-time recurrent neural networks (RNNs). Ensuring the internal stability of the compressed RNN guarantees that of the original RNN. An error bound between the outputs of the compressed RNN and the original one is derived. Moreover, an optimization problem for reducing the bound is formulated, and it is relaxed to a semi-definite programming problem. Furthermore, it is shown that the proposed model compression method produces a compressed RNN whose output is close to that of the original one as a general tendency. The proposed method is demonstrated on a simple numerical example.

This paper proposes an estimation and control method for multi-input-time-delay systems. It has already been established that the model which estimates a single-input-time-delay system increases its order according to the time-delay steps because the model has to store input as states. Therefore, it is possible to estimate time-delay steps by checking the order of the estimated model. On the other hand, the control method for single-input-time-delay systems is known as state predictive control. This paper extends this concept to multi-input-time-delay systems, proposing an estimation method for multi-input-time-delay systems and state predictive control. The effectiveness of this extended approach is validated by simulations.

For water exploration in the lunar polar region, we are considering a path planning method directed towards a predefined and dynamically changing target region (in this case, the sunlit region). We propose a novel path planning method that utilizes a mathematical modeling algorithm for neutron detectors, which is based on a Bayesian network, and adapts Informed RRT^* algorithm. This algorithm automatically sets the search space based on the time-varying environment, such as a changing sunlight region, and incorporates its map cost. We believe that the proposed method can be practically applied for exploration in time-varying sunlight regions and can identify optimal routes. Furthermore, we have applied the proposed algorithm to ideal data and have verified its effectiveness.

Turbochargers and exhaust gas recirculation (EGR) systems are essential devices for improving the combustion efficiency of diesel engines. If these characteristics change, the desired combustion efficiency and exhaust emission performance cannot be achieved. Therefore, there is a need for a mechanism to detect and report changes in characteristics in a timely manner. In this study, we investigated a method for detecting changes in characteristics and identifying the responsible engine device using neural networks (NN) and support vector machines (SVM) from existing sensor information. Under normal operating conditions, where engine characteristics remain unchanged, a NN is developed to estimate various sensor signals and internal signals of the engine control unit (ECU) based on input signals such as throttle valve opening and EGR valve opening. When changes occur in the engine devices, discrepancies arise between the NN's estimated signals and the actual values. To identify the device with altered characteristics, an SVM is constructed based on the estimation errors of the signals. The effectiveness of the proposed method was demonstrated through a simulation using an engine simulator.

Guidance laws, including proportional navigation, have been studied in various fields. However, there are no examples of applying an impact angle control guidance (IACG) law to vertical landing for an unmanned aerial vehicle (UAV) considering the field-of-view (FOV) limitation of the rigid body model. In this study, an IACG law based on proportional navigation is applied to the rigid body problem for a small multicopter-type UAV, and landing simulations using a time-varying asymmetric barrier Lyapunov function are performed while constraining the pitch angle which limitation comes from the orientation of the UAV's sensor.

Trajectory control laws based on neural ordinary differential equations (ODEs) were proposed by the authors in a previous study. In the present study, the trajectory control laws are extended by applying the reservoir computing framework. The parameters that correspond to the intermediate layer of the trajectory control laws using neural ODEs are excluded from the decision variables. This can significantly reduce the number of parameters to be optimized hence the computational cost for training while maintaining the structure of the control laws. This enhancement will allow for use of general-purpose nonlinear optimization algorithms, extending the application of neural ODEs to mission design optimization beyond design of control laws.

This research proposes a hierarchical control method using model predictive control and control barrier functions for CAVs (connected and automated vehicles). By using model predictive control for the upper layer and a control barrier function for the lower one, we aim to achieve control with reduced fuel consumption while ensuring safety even in environments where the behavior of the vehicle ahead is unpredictable. In this study, we aim to improve the feasibility of the lower layer by using Adaptive CBF. The upper layer takes preventive safety actions based on predictions of the preceding vehicle's behavior and ensures feasibility by using slack variables. The lower one using Adaptive CBF ensures the safety that is lacking. The effectiveness of the proposed method is demonstrated through simulations, comparing the results with those of previous studies.

In this paper, we propose a new approach to data-driven parameter tuning method of feedback gain and integral gain for servo system. Differently from other methods, the proposed method does not require reference model to be tracked by the output, which is considered as the key point in the servo system. To realize it, we utilize data-driven prediction, which predicts the response of the closed loop with some controller parameter before experiment. Then, by using such a data-driven prediction, we perform nonlinear optimization under constraint conditions on the predicted input and output. Finally, we show the usefulness of the proposed method with the experimental example.

This paper concerns the realization step of a closed-loop subspace model identification method. Especially, a computation algorithm of estimates of the coefficients (B, D) is discussed. The contribution of the paper is twofold. Inspired by Ikeda's computation algorithm for open-loop PO-MOESP using the oblique projection technique, our proposed algorithm gives smaller estimation error than the conventional one does when the coefficients (B, D) are estimated. Estimation error analysis of the extended observability matrix on our closed-loop identification method is performed. Numerical simulations demonstrate the superiority of the proposed algorithm.

In recent years, there has been an active social implementation of mobile robots mounted 2D LiDAR (2 Dimensional Light Detection And Ranging). When a user remote-controls the robot, it is required to realize collision avoidance with minimum assist input while prioritizing user's input commands as much as possible. This paper adopts control barrier function theory to achieve collision avoidance for a mobile robot. In particular, we propose a new model using the distance obtained from 2D LiDAR and the robot's shape as state variables. The effectiveness of the collision avoidance assist controller designed based on this model is demonstrated through computer simulation.

In this study, we propose a distributed search method for multiple targets in a two-dimensional space using detection maps. The proposed search method has the advantages of being robust to global area coverage, having a small computational load, and being fault-tolerant due to its distributed control. Specifically, each search agent shares its target detection history from its own search as a detection map with other agents, and is controlled according to the gradient of the potential field created based on this map, which enables distributed and efficient search activities. The contribution of this paper is that the proposed method improves the flexibility of conventional search methods by distributing them. Finally, simulation validation confirms the effectiveness of the proposed method.

Numerous studies of brain-computer interface (BCI) attempt to operate something by estimating the brain state. They often try to extract frequency domain features from electroencephalography (EEG). In real-time BCI, bandpass filter (BPF) based feature extraction remains a critical technique. However, due to the low-frequency and narrow-band nature of primary EEG components, the BPF's response delay cannot be ignored. Hence, we propose the spectrum observer that decomposes EEG signals into combinations of sine waves at known frequencies. First, we implemented the spectrum observer as a stationary Kalman filter. Next, we validated that the proposed method achieves feature extraction from raw EEG with the same accuracy as BPF. Finally, response tests with simple signals indicated that the proposed method achieves approximately a 0.1-second reduction in response time compared to BPF.