Various methods have been proposed for analyzing traffic accident hotspots. One of these methods is to detect traffic accident hotspots on the road network using a hypothesis testing method. However, this method does not consider the time of occurrence of a traffic accident. In other words, this method provides information on high-risk locations for traffic accidents but not on the corresponding time. This paper proposes a new method for detecting traffic accident hotspots considering not only location but also time. We therefore extended the previous hypothesis testing method to consider the time of occurrence of a traffic accident. First, we check for changes in spatial properties over time in the target area using local indicators of spatial autocorrelation (LISA) cluster maps. Next, we estimate the probability density function of traffic accidents in the spatio-temporal network using the spatio-temporal network kernel density estimation (STNKDE) method. Finally, we detect clusters where the probability density of traffic accidents is significantly higher through hypothesis testing based on the estimation results. We detect clusters under four conditions while testing different hypotheses and parameters. The results show that the proposed method can detect traffic accident clusters. Therefore, when considering traffic safety measures, the method can be adopted to detect the location and time that require attention and allocate the necessary resources.

The RoboCupRescue Simulation (RRS) is a project that uses multi-agent systems to address real-world problems. RRS needs to perform an optimal task assignment for various disaster relief teams (agents) to minimize the damage caused by disasters. In previous studies, the RRS task assignment problem was modeled as a distributed constraint optimization problem (DCOP) and its effectiveness was confirmed. The DCOP can handle cooperative actions by one type of agent, but it cannot handle cooperative actions by multiple types of agents. To achieve this cooperative behavior, task assignment must take into account the order constraints of multiple types of tasks. It is also difficult for the DCOP to consider the dynamic environment. RRS should consider the dynamic environment because disaster conditions change over time. To consider the dynamic environment, task assignment must take into account the time window constraints of the task. Therefore, the objective of this study is to realize a disaster relief agent that performs task assignment considering the task order and time window constraints for RRS. First, the task assignment problem of RRS was modeled as a layered DCOP (L-DCOP) and L-DCOP was applied to RRS. Next, the rescue abilities and behavior of the L-DCOP agent applying the L-DCOP were compared with those of the DCOP agent applying the DCOP. The results confirmed the effectiveness of the L-DCOP agent and its ability to perform given the constraints of the task.

RoboCupSoccer Simulation 2D(RSS2D) simulates a soccer match between two teams of 11 autonomous mobile robots each on a virtual two-dimensional plane. In soccer, the ball and players are always on the move, leading to a constantly changing game situation. RSS2D agents thus need to constantly analyze the match situation and predict the actions of enemy agents in determining their actions. However, it is not easy to analyze a player’s actions, because it is necessary to combine specialized knowledge with consideration of the intentions of all players’ actions. In human soccer, an analysis method using the expected possession value (EPV) has been proposed to solve this problem. The EPV is a measure of the probability that a team will score or concede a goal according to the game conditions and player actions. In this study, we apply the EPV to RSS2D to analyze match situations and agent actions. We also examine whether EPV-based analysis is effective in an RSS2D environment. We create a dataset by extracting agent actions and ball position information from RSS2D log files. Next, using the dataset as input, we train several estimation models using a convolutional neural network. Then, by combining the multiple estimation models, we create an estimation model for the EPV of passing. The effectiveness of the estimation model for the EPV of passing is confirmed by analyzing game conditions and agent actions.

Recently, research on AIWolf, which is artificial intelligence that plays the Werewolf game, has attracted much attention. Werewolves in the Werewolf game should cooperate using “whispers;” which are conversations among werewolves only. However, whispers have not been used effectively in the AIWolf Competition and little research has been conducted on this topic. Therefore, the purpose of this study is to clarify the effectiveness of strategies that use whispers in AIWolf. A wide variety of strategies use whispers, but we focus on unifying attack targets, which was shown to improve the winning rate in previous studies. In the strategy of unifying attack targets, the strategy of the guiding agents is important. Therefore, we used 13 AIWolf Competition finalists with a wide variety of strategies and added statements of whom to attack in whispers to each finalist as a guide for whom to attack. Because there are three werewolf competitions in the AIWolf Competition setting, we created two agents to be attuned to the attack target of one of the guiding agents. In an experiment, we used these agents in the same setting as in the AIWolf Competition, and analyzed the differences between when the attack targets were unified and when they were not. The results showed that when the werewolves were a combination of two attuning agents and one guiding agent, the werewolves’ win rate increased by an average of 2.3 percentage points. We also found that attuning to the agent with a high winning rate or high success rate in attacks improved the winning rate. These results indicate that attuning only the attack targets of agents with high winning rates and high success rates in attacks led to higher winning rates.

In Japan, police and researchers work together to reduce the incidence of road traffic accidents by analyzing space and time information such as the occurrence of traffic accidents and traffic volumes. Traffic safety studies have analyzed data on traffic accidents adopting various approaches to reduce the number of traffic accidents. However, there is no common database that can be used in such studies, because each study has a different focus of analysis and different granularity of the geographic area. Here we show a database for traffic accident analysis that we designed based on the requirements of previous traffic safety studies. Traffic accidents are spatiotemporal events. Therefore, the database requires the manipulation of spatiotemporal data. In this paper, we represent accident situations by geospatial objects to enable manipulation of the spatiotemporal data. The designed database maintains information as a geospatial object. All non-spatial data depend on spatial data according to the design of the relation. The database combines tables representing traffic accident information using spatial predicates defined in Simple Feature Access. Thus, the database provides capabilities for the manipulation of any items necessary to analyze accident situations spatiotemporally. The study developed a database and scripts that register the data and analyze the data to evaluate the designed database. The results show that the designed database has good operational performance in terms of scalability and efficiency. The designed database is thus useful for the analysis of traffic accident data.

In the recent years, the evolution of artificial intelligence (AI) has influenced the development of human-interactive communication games. The Werewolf game was originally a real-world, face-to-face, indoor game played between a minimum of four players. In this research, we focus on Werewolf BBS, which is a web-based game, based on the real-life Werewolf game. In Werewolf BBS, players can send various messages among themselves, including a wide variety of chat messages and discussions on current affairs. Thus, it is challenging for AI to understand the phrases being passed on by players. The aim of this study is to understand the content of the messages being communicated in Werewolf BBS. In this paper, we have introduce an automatic classification method based on machine learning to categorize the phrases into six groups. Previous studies have used unavailable information that must be excluded when introducing AI to the game. Thus, we hereby propose a method to classify phrases without additional information using a support vector machine (SVM). SVMs are discriminative classifier models of machine learning. In this paper, an SVM is used as a tool for categorization; it discriminates the different expressions shared by the players into the six groups. The results of an experiment confirm that the messages can be classified without preconditions. This verifies the effectiveness of the present research.