|
A Target Assignment and Path Planning Framework for Multi-USV Defensive Pursuit |
|
Minjoon Lee(KAIST, Korea), Min Kyu Shin(Korea Advanced Institute of Science and Technology, Korea), Il-Chul Moon(KAIST, Korea), Han-Lim Choi(Korea Advanced Institute of Science and Technology, Korea) |
| This paper presents a target assignment and path planning framework for multi-USV defensive pursuit. The assignment is made by using Proportional Navigational guidance(PNG) law to precisely calculate the Predicted-Impact-Point(PIP). The paths to the targets are generated by dubins path and iterative Linear-Quadratic-Regulator(iLQR) method to avoid the collision. Simulations on combat situation with swarm USVs demonstrate that the assignment is made based on PIP, and the path is generated while detouring around the obstacle. |
|
|
End-to-End Control of USV Swarm Using Graph Centric Multi-agent Reinforcement Learning |
|
Kanghoon Lee, Kyuree Ahn, Jinkyoo Park(KAIST, Korea) |
| This study proposes an algorithm to derive the decentralised and cooperative control strategy for the USV swarm using graph centric multi-agent reinforcement learning (MARL). The model first expresses the mission situation using a graph considering the various sensor ranges. Each USV agent encodes observed information into localized embedding and then derives coordinated action through communication with the surrounding agent. To derive a cooperative policy, we trained each agent’s policy to maximize the team reward. Using the modified prey-predator environment of OpenAI gym, we have analyzed the effect of each component of the proposed model (state embedding, communication, and team reward) |
|
|
Development of a Robust Path Following Controller for an Unmanned Surface Vessel |
|
Jongho Shin(Chungbuk National University, Korea), Hyeon Kyu Yoon(Changwon National University, Korea) |
| This paper proposes a robust path following control method for an unmanned surface vessel (USV). In order to do so, a nonlinear dynamic model of the USA is developed at first and the nonlinear system is rearranged as a linear model structure. With the derived linear model, a feedback linearization-based controller is developed assuming that the linear system can reflect the overall nonlinear USV model. For compensating the discrepancy between the linear and nonlinear models, this study utilizes the disturbance observer (DOB). To validate the performance of the proposed method, path following simulations are performed and the results are analyzed. |
|
|
Hierarchical Task Planning Considering Communication Status for Multi-Robot System |
|
Jiyoun Moon(Chosun University, Korea) |
| As the roles of robots have expanded, flexible task planning frameworks have received significant attention in various domains. In particular, task planning based on multi-agent systems is essential because tasks can be performed efficiently in broad areas by utilizing such systems. This paper proposes various task-planning methods according to the communication statuses of robots. The proposed methods were successfully verified through experiments in which a multi-agent system consisting of a central control system and several robots was applied. Our experimental setup assumed that the central control system and robots had different communication statuses. |
|
|
Curriculum-Based MADDPG for Defensive Task Assignments |
|
Minhyuk Yoon, J.Hyeon Park, Dabin Kim, H. Jin Kim(Seoul National University, Korea) |
| When multi agent reinforcement learning(MARL) is applied to military environments, scalability to number of agents is essential because they involve large number of populations. However, most of the MARL algorithms do not satisfy scalability. Therefore, we utilize curriculum learning-based MARL on our setting. In curriculum learning, we divide MARL into stages and double the number of agents on each stage. We use the evolutionary algorithm, by treating each agent sets as genes and progressively evolving them. We compare our implementation and vanilla MARL on four and eight agents in the task assignment scenario. As a result, we confirm the improvement of the performance over vanilla MADDPG. |
|
|
Design Considerations of Real-time Radar Sensor Modeling for Unmanned Surface Vehicle (USV) |
|
Hyeonseok Lee, Hyeonhee Yi, Jung-Dong Park(Dongguk University, Korea) |
| We present a design of the real-time radar sensor model for unmanned surface vehicles (USV). To construct an efficient learning environment of an unmanned surface vehicle (USV) for the swarm operation, accurate virtual modeling of the radar sensor with a light processing load is necessary. To achieve real-time modeling of the marine radar operations with a high level of modeling accuracy under a limited computational power, our work is to extract the signal-to-clutter noise ratio (SCNR) by considering physical radar specifications with pre-extracted target radar cross-section (RCS) using a 3D-EM simulator (HFSS). Modeling of various clutters such as rain, snow, fog as well as sea clutter has |
|