|
Formation Path Learning for Cooperative Transportation of Multiple Robots using MADDPG |
|
Kenta Miyazaki, Nobutomo Matsunaga, Kazuhi Murata(Kumamoto University, Japan) |
| The cooperative transportation using multiple robots has been expected to be used in factories and construction sites. Cooperative transportation can support various situations compared to transportation using one robot. However, the industrial application of cooperative transportation has not been advanced due to the complexity of formation change. In this paper, the formation change using multi-agent deep deterministic policy gradient is proposed, which is a deep reinforcement learning method specialized for multi-agent systems. The effectiveness of method is evaluated by simulations.
|
|
|
A RGB-D Vision based Indoor SLAM using 2.5D Map by Multiple UAVs |
|
Hyunseung Kang(Korea Polytechnic University, Korea), Kyomun Ku, Jaehong Shim(University, Korea) |
| This paper presents an approach to build an indoor 2.5D map with multiple UAVs(Unmanned Aerial Vehicles). In indoor environment, GPS system would be denied so each UAV adopted tracking camera to localize itself and 3D data is acquired by stereo depth camera to build a map. However its raw data can contain some noise and have large data size, octree filter is applied. Each UAV can build a map for part of the global floor and each 3D local maps are converted to 2.5D then merged together. To make a global map from numerous local maps, feature points are extracted that can be detected in 2.5D of the indoor environment. |
|
|
Goal Assignment in Leader-Following Formation Control of Second-Order Multi-Agent Systems |
|
Yun Ho Choi(Korea Institute of Science and Technology (KIST), Korea), Doik Kim(Korea Institute of Science and Technology, Korea) |
| In this paper, the goal assignment problem of second-order multi-agent systems is considered in the leader-following formation control framework. The goals of followers are assumed to be interchangeable. During the operation of the multi-agent systems, the goals of a pair of followers are exchanged locally if an assignment condition is satisfied. Compared with the existing goal assignment studies utilizing the position information, the main feature of this paper is to derive a velocity-based assignment condition to deal with the assignment problem of second-order multi-agent systems. |
|
|
Behavior Tree Driven Multi-mobile Robots via Data Distribution Service (DDS) |
|
Seungwoo Jeong, Taekwon Ga(Yonsei University, Korea), Inhwan Jeong(Hyundai Robotics, Korea), Jongeun Choi(Yonsei University, Korea) |
| We propose multi-task programming for multi-robots via a behavior tree (BT) as a mean to reduce programming
complexity of multi-mobile robots. For multi-mobile concurrent control, robot sensors or actuators data are shared via data distribution service (DDS). In addition to the advantages of DDS, the BT is deployed to control and monitor multi-mobile robots, The BTs are inherently modularized and reactive to the environments, which enables a flexible utilization of multi-mobile robots. With the help of Robot Operating System (ROS2), The BTs are implemented with
ROS2 nodes on the DDS that can be executed in both servers and clients for the task allocation. |
|