Mobile manipulators are widely used in manufacturing industries. This paper introduces a hybrid navigation method combining A* and Q-learning algorithms for a mobile robot equipped with a 6-DOF manipulator arm for pick and place operation. The robot autonomously navigates in the environments by scanning surroundings with laser scanners and mapping in ROS, enabling obstacle avoidance and trajectory optimization. The A* algorithm handles global path planning, generating optimal routes, while Q-learning manages local planning by adapting to real-time changes. Results demonstrates environment mapping in RViz with the robot's ability to navigate from start to goal state and the advantages of the hybrid approach in partially or fully known static environments.

Japan has experienced several significant earthquakes in the past that have caused extensive damage. Moreover, large earthquakes are predicted to occur in the future. In our previous study, we introduced a time-series model of the number of evacuees. We classified evacuation factors into four categories and derived a dynamic model of the number of evacuees based on the classification. This kind of model is useful for planning evacuation center operations and supporting victims. As an extension of the previous study, we construct a model that considers the shortage rate of supplies based on the recovery rate of lifelines, such as electricity and roads. The model explicitly considers the capacity gap between potential evacuees and actual shelter evacuees, which varies significantly across regions. In addition, the impact of the recovery rate of lifelines and the degree of stockpiling on the shortage rate of supplies was examined through simulations. The proposed model can reflect the detailed damage situation in the affected areas and predict the number of evacuees more realistically.

This paper considers a distributed model predictive control problem for multi-agent systems. Without an event-triggered mechanism, the conventional alternating direction method of multipliers (ADMM), which requires frequent exchange of information between agents, can give an optimal solution to the problem after several tens of iterations. With an event-triggered mechanism, the communication-censored ADMM (COCA), which allows restricted information exchanges, can give an optimal solution. This paper applies the COCA to the problem above with the event-triggered mechanism, especially in a leader-follower setting. The numerical examples for single and double integrators illustrate the effectiveness of the COCA for distributed model predictive control.