The video of investigation inside the primary containment vessel (PCV) of the Unit 2 of the Fukushima Dai-1 number power plant in 2019 is too long with a lot of redundant information and has a lot of focus/lighting problems. This study, in order to enhance convenience for analyzing and utilizing the video inside the PCV, attempts to implement the following modules: 1) reduction of video time (time-shortening processing), 2) extraction of impressive scenes (digest image creation), and 3) generation of wide-area (panorama) images.

The conventional multivariate analysis method using chemical data obtained by elemental analysis has been still challenging to discriminate materials are included in composite materials. Hierarchical cluster analysis (HCA) and Dempster–Shafer theory, that are series of multivariate analysis method, can evaluate the similarity of each material using the chemical data; however, are still difficult to identify the materials contained in a composite material. In this study, evaluating the similarity of elemental analysis data by HCA was conducted, and then stepwisely evaluating was implemented using degree of belief. In other words, it is a stepwise identification means of composite materials that combines the distance output from HCA with a concept of the degree of belief. Artificial model samples were demonstrated implementing in the result of standard reference materials such as metal alloys and plastics. Consequently, the proposed method consistently results prerequisites, and the degrees of belief were quantitatively calculated for each sample confirming the usefulness of this method.

Rough Set Non-deterministic Information Analysis (RNIA) is a mathematical framework for analyzing categorical tabular data and can be considered as a framework that adds information incompleteness to rough set theory. In RNIA, the DIS-Apriori method generates rules from the regular table Deterministic Information System (DIS)， and the NIS-Apriori method generates certain rules and possible rules from the table Non-deterministic Information System (NIS) with missing values or non-deterministic information.

In this paper, we add two new features, missing value imputation and machine learning, to RNIA. It is usually considered difficult to generate rules from table data sets containing many missing values, however, we can generate certainty rules from such tables. Using this property, we generate certain rules with an attribute including missing values as the decision attribute and apply them to impute missing values. We term this procedure Self-obtained Rule-based Imputation procedure (SRI method) for imputing missing values. This method is unsupervised learning and does not require background knowledge or additional information. In addition, if the SRI method is applied to each attribute including missing values, it is possible to realize the functionality of self-learning DIS sequentially from NIS. We term this framework Machine Learning by Rule Generation (MLRG). Through some experiments, we clarify the SRI method’s validity and consider the possibility of MLRG framework.

Hierarchical reinforcement learning (HRL) is an approach that incorporates intrinsic motivation mechanisms into reinforcement learning. HRL divides the agent’s internal mechanism into two components: a higher-level policy (application order of subgoals) and a lower-level policy (behavior sequence to subgoals) for problem solving. It has been demonstrated that HRL can solve problems in environments with sparse rewards and environments that require learning of long action sequences, which are difficult to address with conventional reinforcement learning, provided that the definition of subgoals is appropriate. However, existing HRL assumes the availability of predefined subgoals necessary for problem solving and does not provide an algorithm for achieving autonomous reinforcement learning. In this study, we propose stepwise unified hierarchical reinforcement learning (SUHRL), a new reinforcement learning algorithm that introduces a mechanism to gradually generate necessary experiences and appropriate subgoals for problem solving. SUHRL solves problems by stepwise clustering using Fuzzy ART and experience acquisition processing to generate suitable subgoals incrementally. Evaluation experiments conducted on MiniGrid environments and Montezuma’s Revenge demonstrate that the proposed method can generate the required subgoals incrementally and achieve autonomous problem solving.

In this study, we focus on clothing as a design aspect of robots, and investigate users’ impressions when robots change their clothing to suit the task at hand. Clothing can have an effect of showing one’s role and abilities to others. By utilizing this effect in the appearance of robots, it may be possible to express multiple roles and abilities for different tasks with a single robot, which could be beneficial in robot design. In the experiment, we investigated the difference in user impressions when the robot changed its clothing to suit the task at hand compared to when it did not. This was done through a video-watching experiment conducted on a simulator. The results of the experiment showed that when the robot changed its clothing to suit the task at hand, users perceived its appearance as appropriate for the task. This suggests a potential for effectively conveying the robot’s abilities and characteristics to users through its clothing.