HAMT 2019 Abstracts


Short Papers
Paper Nr: 1
Title:

Multi-Agent Base Evacuation Support System Considering Altitude

Authors:

Shohei Taga, Tomofumi Matsuzawa, Munehiro Takimoto and Yasushi Kambayashi

Abstract: In this paper, we propose an extension of an evacuation support system that we have previously proposed (Taga, 2018). The system suggests evacuation routes in cases of disasters. We have confirmed the usefulness of the system. When a disaster occurs, we anticipate that the current popular wireless communication based on the Internet may not be very reliable. In order to accommodate such a problem, our proposed system employs multiple mobile agents and static agents on smartphones that use a mobile ad hoc network (MANET). The proposed system collects information by mobile agents as well as diffuses information by mobile agents so that the system provides an optimal evacuation route for each user in a dynamically changing disaster environment. In simulations, our system successfully guides evacuation users to safe areas. The system, however, does not consider the altitude of the evacuation routes. Therefore, the system may not be very useful in cases of flood. When a tsunami or a flood tide occurs, low altitude location may be under water. Therefore, evacuees need to move along high altitude routes. In this paper, we also take account of the altitude information for constructing evacuation routes.

Paper Nr: 2
Title:

Development of Agents for Creating Melodies and Investigation of Interaction between the Agents

Authors:

Hidefumi Ohmura, Takuro Shibayama, Keiji Hirata and Satoshi Tojo

Abstract: In this study, we attempted to construct computational musical theory by creating musical structure using physical features of sound without relying on the existing musical theory. Subsequently, we developed an agent system to create melodies. The agents can select the next note (a sound timing and a pitch) depending on the lattice spaces consisting of physical relationships (ratios) and probabilities. Further, we improve the agents which are interacting with each other in the system, and the system outputs various sounds such as music. We confirmed that the system could create structures of musical theory, such as mode, scale, and rhythm. The advantages and disadvantages of the lattice spaces are discovered.

Paper Nr: 3
Title:

Development of Agent System for Multi-robot Search

Authors:

Masashi Omiya, Munehiro Takimoto and Yasushi Kambayashi

Abstract: In this paper, we propose an agent system that controls multiple mobile robots. We describe the agent system as well as an example multi-robot system as an application of this agent system. The aim of the multi-robot system is providing efficient searches for a given target. Therefore, it is necessary to mutually communicate and cooperate among robots. Taking account of the delay of communication, cost, fault tolerance, and robustness, we have chosen mobile agent system for the information transmission method. We have also chosen ad-hoc method for the communication mode, where each robot communicates directly without going through the Internet. Even though agent systems are often implemented using Java language or Python language, we have chosen C++ language for developing our agent system with the hope for gaining performance efficiency. We have modularized each function and it does not depend on other modules. We have shown the feasibility of our multi-agent system by applying the system to a multi-robot system that implement particle swarm optimization.

Paper Nr: 4
Title:

Predicting Fault Proneness of Programs with CNN

Authors:

Kazuhiko Ogawa and Takako Nakatani

Abstract: There has been a lot of research aimed at improving the quality of software systems. Conventional methods do have the ability to evaluate the quality of software systems with regard to software metrics. (i.e. complexity, usability, modifiability, etc.) In this paper, we apply one of the deep learning techniques, CNN (Convolutional Neural Network), in order to infer the fault proneness of a program. The CNN approach consists of three steps: training, verification of the learning quality, and application. In the first step, in order to make training data, we transformed 27 program source codes into 1490 images with colored elements, so that the features of the images remain. In the second step, we set the goal of the accuracy of machine learning and trained with the training data. In the third step, we forced the trained system to infer the fault proneness of 692 images, which were transformed through 5 programs. This paper presents the effectiveness of our approach for improving the quality of software systems.