The following 6 works have been accepted for presentation.
|Exploration of disaster scene with a SPOT robot
|Keisuke Ando, Haruaki Ueda, Takeshi Uchitane, Kazunori Iwata, and Nobuhiro Ito
|Applying covariance structure analysis to reveal relationships between agent capability and map characteristics in RoboCup Rescue Simulation
|Akira Hasegawa, Yuki Miyamoto, Haruki Uehara, Takeshi Uchitane, Kazunori Iwata, and Nobuhiro Ito
|A comprehensive simulation management platform for RRS on public cloud
|Arman Absalan and Amir Aslan Aslani
|RoboCup Rescue Simulation Web Viewer
|Gabriel Amorim, Bruno Rego, and Guilherme Miazaqui
|Disaster management: A multiagent system based approach
|Vinaysheel K. Wagh, Pramod Pathak, Paul Stynes, and Luis G. Nardin
|An evacuation route model for disaster affected areas
Exploration of disaster scene with a SPOT robot
When a building is on fire, the SPOT robot is a very versatile robot to explore the rooms for victims. During its exploration, both a metric and topological map of the building has to be made. For transitions from the corridor into the rooms the doorways have to be recognized, with define gateway points. For the topological map it is also important to distinguish the different rooms, it has to be clear if one is in the kitchen or in one of the bedrooms.
To enhance the situation awareness of the robot a 360 camera is mounted on the robot, which provides a panoramic overview of the surroundings, including a depth map and potentially a full 3D reconstruction of the rooms, including semantic information. The semantic information can be used to feed a knowledge graph with high-level information.
Applying covariance structure analysis to reveal relationships between agent capability and map characteristics in RoboCup Rescue Simulation
Keisuke Ando1, Haruaki Ueda1, Takeshi Uchitane1, Kazunori Iwata2, and Nobuhiro Ito1
This study discovered environmental factors important in disaster rescue operations of RoboCup Rescue Simulation (RCRS), a large-scale disaster rescue simulation. All environmental factors are defined from the map characteristic indices that represent city structure complexity. Moreover, these factors are constructed by applying factor analysis to the dataset of these indices calculated from real-world maps. Therefore, the discovered environmental factors can interpret as intervening between the map characteristic indices and the results of rescue operations that indicate the agent team’s capability. This result enabled to evaluation and compared agent teams in different environments based on environmental factors.
A multi-agent simulation derives results from a pre-built agent algorithm and environmental information that affects agents’ behavior. In the same way, in a disaster rescue simulation, the agent team’s rescue strategy and the characteristics of the rescue target area lead to the results of the rescue operation. The final objective of this study is to clarify the relationship in this derivation. Our previous studies currently find out some relationships.
In those studies, we quantified indices of city structure complexity as map characteristics obtained from the map structures. We estimate the relationship between map characteristics and the results of rescue operations using generalized linear mixed models and kernel regression analysis. Based on the estimated relationship, it was possible to explain the behavior of each agent team from map characteristics for several agent teams. However, we cannot use this method to compare multiple agent teams.
This presentation presents a new approach introducing covariance structure analysis to compare multiple agent teams in this relationship. This approach first defines environmental factors of rescue operations by performing factor analysis on map characteristics. Then, through the defined factors, we construct a model that assumes a relationship between map characteristics and the score representing the results of rescue operations in RCRS. Furthermore, we apply covariance structure analysis to evaluate each agent team’s capability based on three representative agent team’s algorithms in RCRS and the constructed model.
The analysis results show that the approach enables the comparison of rescue operations among multiple agent teams in detail. In addition, due to the properties of the covariance structure analysis, we were able to graphically represent the relationship between map characteristics and the results of rescue operations. The graphical models could explain the relationship based on the behaviors of the agent teams.
A comprehensive simulation management platform for RRS on public cloud
Akira Hasegawa1, Yuki Miyamoto1, Haruki Uehara1, Takeshi Uchitane1, Kazunori Iwata2, and Nobuhiro Ito1
A disaster relief simulation requires a large number of runs with various parameters in general. In other words, researchers are required to spend a lot of time on complex preparation and execution management for the simulations. This presentation proposes a simulation platform on cloud computing that enables easy and flexible execution for RoboCup Rescue Simulation (RRS). This platform provides RRS researchers with a remote management feature for large-scale simulation execution. It allows researchers to use more time on developing and improving algorithms, which they should focus on. Therefore, this platform can be a powerful tool for natural disasters.
The Great Hanshin earthquake is an urban earthquake, which occurred in 1995 and killed 6434 people. The RoboCup Rescue Simulation project has started in the wake of the earthquake and has developed a multi-agent simulator. The simulator represents disaster situations such as fires, collapsing buildings, debris, and injured civilians. In addition, it simulates the rescue operations by the multiple agents for the disasters.
Researchers need to find meaningful results, that may be AI, robotics, and natural disaster, through many simulations under various conditions with realistic possible conditions. It requires large-scale computing resources to achieve it. In addition, preparing the execution environment for large-scale computer simulations is complex. Furthermore, it is complicated to manage execution under multiple conditions, monitor the execution status, and manage the results. Thus, researchers require computing resources that are flexible and easy to prepare. Moreover, they require execution environments that are easy to manage, too.
This presentation proposes a design and implementation for a comprehensive simulation management platform for RRS on public cloud. We achieved this by using Docker to virtualize the execution environment of RRS as containers, Amazon Elastic Container Service to run the virtualized containers on AWS, and Cloud Formation for simulation scalability, AWS Batch for multiple simulation executions with various parameters As described above, we have made it easier to procure and expand computing resources than a simulation environment using desktop computers. We also prepared a web interface to upload agent programs and management execution. This interface allows new researchers to use a simulation environment without tangled preparations.
We confirmed that the platform is effective through about 100 simulations in RoboCup JapanOpen 2020, an online competition of RRS.
RoboCup Rescue Simulation Web Viewer
Arman Absalan and Amir Aslan Aslani
In order to show Rescue Simulation competitions online, a service called RCRS-Web Viewer has been designed and operated. This web-based service would gather competition logs from RCRS-servers and serve them on the web. The RCRS-Web Viewer is composed of two sections. Viewer and back-end. Gathering and serving the logs is related to the back-end part and the Viewer by loading a log, will display the competition. The aim of this presentation is to introduce different parts of the system and indicating some complexity.
Disaster Management: A multiagent system based approach
Gabriel Amorim, Bruno Rego, and Guilherme Miazaqui
One of the possible approaches that is addressed in this work is the partitioning of the exploration region using the K-Means clustering algorithm. With the motivation of maximizing the quantity of civilians saved and improving agents’ efficiency, the proposed architecture implements centralized clustering at the beginning of the simulation considering the area of the buildings in each region. Furthermore, centralizing commands was proposed in opposition to standard autonomous behavior of agents. This proposal was implemented in two different levels, where in the first only target allocation would be made, while in the second exploring the buildings would also be commanded in addition to the target allocation. To evaluate the results, 450 data collections were carried out spread across five different simulation maps, seeking to measure the impacts provided by the proposed strategies. While the main utilized metric was the final score provided by the simulator itself, others were used such as the number of rescued civilians and the number of identified civilians. It was concluded that, although a statistically significant difference was found in the final score in one of the maps for different K’s, the gain achieved from organizing actions in a centralized manner was higher.
An evacuation route model for disaster affected areas
Vinaysheel K. Wagh, Pramod Pathak, Paul Stynes, and Luis G. Nardin
Natural disasters such as earthquake severely damage buildings and introduce obstacles to people trying to evacuate an affected area. Detecting and analyzing the severity of damage to an affected area is a challenge. This paper proposes a novel model for classifying damaged buildings and supporting people’s evacuation from natural disaster affected areas using satellite images. The model integrates image segmentation and classification with a shortest path algorithm. First, buildings are detected from pre-disaster satellite images using the proposed Segmentation model. Second, post-disaster images are classified based on the severity of the damage using the proposed Classification model. Finally, the shortest and safest evacuation route to a rescue shelter is detected using the Dijkstra’s algorithm. Results show that the Route Detection model dynamically adapts to new and updated satellite images. The Segmentation model shows an F1 score 5% better than the Building Footprint Extraction model and the Classification model shows F1 scores 8% and 10% better than the VGG16 and VGG19 respectively. The Evacuation Route model is useful to disaster management teams and trapped people for planning safe evacuation routes out of the affected area.