We propose a hammering system which is small and light-weight to mount on an UAV and a control method for stable contact to the infrastructure wall for the inspection. In Japan, all of the infrastructure must be inspected for each five years. However, the inspection of high place is dangerous, so inspection robots are needed. Our research group studied an investigation using UAV. In this research work, we realize small and light-weight hammering system for the inspection by UAV, and control method for stable contact between the hammering system and the inspection wall. This paper shows the design of the hammering system, the evaluation of the soundproof and contact force control of UAV and inspection wall.

In this paper, we propose a Q-learning method by using dual Q-table. Concretely, the proposed method has two Q-tables: “Whole Q-table'' is larger, based on the whole space (in enough detail for learning optimal actions) of the environment and “Partial Q-table'' is smaller, based on a subspace (rough for learning outline actions) of the whole space. The two Q-tables simultaneously learn the environment based on a selected action. The action is selected by using the more learned Q-table out of the two Q-tables by each step. We simulated the proposed method, comparing with conventional ones, under three conditions of the learning environments, where the partial Q-table can learn optimal actions at the highest rate, middle rate and the lowest rate of the situations in the environment. As a result, we indicated that the proposed method can learn the optimal actions at any rates. As the rate is higher, it converges earlier. Even if at the lowest rate, the proposed method is almost as effective as conventional one. And we indicated the proposed method to be effective by using mathematical analysis. Furthermore, we verified that the proposed method was effective under an actual environment.

In this research, human behavior during dialogue is modeled, with the goal of generating human-like motion for humanoid robots. Most of the previous studies on the human motion modeling aimed to model the single motion of human [8] and produce complicated behaviors by combining the models of the single motions [11]. In these studies, each model was associated with a label such as waving a hand, bowing and so on. However, since the human-like behavior in a daily life is diverse and ambiguous, defining a clear label for each motion of such behavior is difficult. To treat with this problem, in this paper, we collected the human motion data during a dialogue and propose a modeling method using Generative Adversarial Networks (GAN) [15]. The result of the human-robot experiment based on the subjective evaluation of the participants suggested that the human-like behavior of the robot could be generated by the proposed method.

This paper proposes a new path planning architecture with consideration of motion uncertainty for wheeled robots in rough terrain with multiple types of terrain. The proposed scheme uses particles to describe the uncertainty propagation between different terrain types and to expand RRT (Rapidly-exploring Random Tree) based on the uncertainty of each node in order to prevent the expansion of position uncertainty. The effectiveness of the proposed method is evaluated on the simulation which hypothesizes Mars enviroment. The simulation results show that the proposed method can prevent the expansion of position uncertainty substantially, which can decrease the times of path following or re-planning.