This paper addresses the design of an assistance control system that enables safe and flexible operation while facilitating operator upskilling. A key factor in upskilling is increasing opportunities for operators to attempt control actions. To this end, we design an assistance system that reduces intervention frequency and promotes operator freedom. We particularly focus on designing a human assistance system for water level control in a tank system. Then, we characterize the set of admissible control actions for the human operator in order to design assistance control logic that ensures the safety of the overall system. Finally, we conduct a human-in-the-loop simulation of the tank system. Through the simulation, we verify that the proposed assistance system ensures the safety of the overall system with minimal intervention, demonstrating its potential as a foundational technology for effective operator upskilling.

Virtual Power Plants (VPPs) have been increasingly being used to achieve carbon neutrality in energy systems and to improve resilience. This study handled a control scheme for small-scale community-operated VPPs, which has attracted much attention in recent years. In such community-operated VPPs, operations are not limited to maximizing economic value, but are also focused on community preferences. In this study we proposed Model Predictive Allocation Control (MPAC), which enables VPPs to operate in a way that appropriately reflects community preferences. The MPAC formulates the energy resource allocation control by Model Predictive Control (MPC) and modifies the VPP operation by tuning the weight parameters of the MPC. Furthermore, the weight parameters can be tuned by a preference learning-based optimization algorithm to easily reflect the community’s decisions. We also compensate for the operational stability of the VPPs by using frequency stabilizing control in combination. The effectiveness of the proposed method was verified by experiments using numerical simulations.

This study examines how to properly control each autonomous machine at a construction site where human-driven construction machines, an autonomous excavator, and an autonomous truck are coexisting. We proposed to utilize Model Predictive Control (MPC) to control two different control targets, an excavator and a truck, using the same method in a unified manner. By taking advantage of the MPC's ability to handle constraint conditions explicitly, we proposed a control method that allows each moving object to operate safely without contact. In order to control each autonomous machine with MPC, the future behaviors of machines other than itself are needed. In this study, a simple but effective model was used to make predictions for human-operated construction machinery. On the other hand, for autonomous machines, we proposed a method to achieve efficient cooperative behavior by utilizing the calculation results of the MPC of other machine. The effectiveness of the proposed method was confirmed by numerical simulations.

Recently, a technique called data-driven control, in which controller parameters are tuned directly using data, has attracted much attention. Among the data-driven control methods, the methods called Virtual Reference Feedback Tuning (VRFT), Fictitious Reference Iterative Tuning (FRIT), and Estimated Response Iterative Tuning (ERIT) have the advantage of allowing parameter tuning with only one-short experimental data. Whether VRFT, FRIT, or ERIT, the data-driven controls are applied, we need to design the reference model appropriately. Selecting an appropriate reference model may require trial and error as well as tuning control parameters. To solve this problem, some previous studies have proposed methods in which the reference model itself also has tuning parameters, and the parameters are automatically tuned to minimize a specific evaluation metric. However, this specific evaluation metric may not adequately represent the response that the system operator wishes to achieve. In particular, if the response that the system operator wishes to achieve cannot be quantified, it is difficult to design a specific evaluation metric. In this study, we propose a method for designing a reference model that focuses on the responses that the system operator wishes to achieve. Our proposed method combines the data-driven prediction used in ERIT with the preference-learning framework to efficiently design a reference model. We confirmed the validity of the proposed method by numerical simulation.

With the development of the electronic commerce business, the need for automation of warehouse transport is increasing, and there are high expectations for the improvement of warehouse efficiency with automated transport robots. Conventionally, such systems have often used path planning for each robot based on a predefined graph-based map, and guiding the robots to their destinations by preventing collisions through area exclusive control or other means. However, conventional methods require precise map information of the warehouse for path planning with high degree of freedom. This makes it time-consuming to update the map information when the layout of the warehouse is changed. In this study, path planning with a high degree of freedom for multiple robots in a warehouse is achieved through model predictive control using map information which is easy to maintain. The proposed method uses simple global map and obstacle map to achieve efficient path planning that takes collision avoidance into account for the robots. Numerical simulations are conducted to verify the effectiveness of the proposed method for improving travel efficiency.

With the development of the Electronic Commerce (EC) business, the need for automation of warehouse transport is increasing, and transport robots are being introduced in various warehouses. This study addresses the problem of generating a path for multiple robots operating in a warehouse to move from arbitrary initial positions to target positions. To optimize the efficiency of all robots’ movements without collisions in a warehouse, we propose a path generation method using Model Predictive Control (MPC). In actual operation, it is important to generate paths that take into account the uncertainty of each robot’s self-position measurement and consider safety margins so that robots do not collide with each other. In this study, we devise two methods for considering safety margins in MPC: penalty functions and constraints, and evaluate the advantages and disadvantages of these methods.

Shaking tables are test devices to check the seismic strength of a structure. If the control performance of the shaking table is poor, the desired seismic waveform cannot be reproduced, and the test desired by the user cannot be carried out. Control adjustment by the control operator is necessary to ensure that the shaking table provides adequate control performance. However, the control adjustment is very difficult for unskilled operators because the response of the shaking table changes due to the influence of the reaction force and nonlinear friction of the specimen. We expect that the application of the data-driven control will make it possible for unskilled operators to easily design the optimal control parameters. In this study, we added two feedback controls to suppress the effects of the reaction force and nonlinear friction, respectively. Then, we applied data-driven control to adjust feedback control parameters. We verified the effectiveness of the proposed method through numerical simulations.

This paper addresses robust filtering for nonlinear systems with parameter uncertainties. We developed a new robust unscented Kalman filter (RUKF) which doesn't require calculating Jacobian matrix by using Unscented Statistical Linearization to consider the influence of parameter uncertainties of covariance matrices. The RUKF is more accurate than conventional UKF when the systems have parameter uncertainties. However, when there is no parameter uncertainty, estimation accuracy of the RUKF may be inferior to that of the UKF. Then, we also developed adaptive RUKF (ARUKF) by introducing an adaptive scheme into RUKF to automatically tune the influence of parameter uncertainties. The validity of the proposed methods is illustrated by Monte Carlo simulations.