A Study for Development of Autonomous Paddy-weeding Robot System -An Experimentation for Autonomously Workspace-Cognition and Counter-Rotation Turn-

In the agricultural field of Japan, there is two issues are facing; population aging and labor shortage are progressing, rapidly. In order to solve these problems, it is indispensable to create a method that improves productivity and labor-saving technology in agricultural works. A weeding robot is one of an approach for achievement of these problems, it can be considered that labors will be released from tough or dangerous works. In this study, autonomously controlling the weeding robot will be focused on. To achieve this, an algorithm will be proposed that autonomously constructs a state-action space based on various sensor information and can apply it to actually work. The main contribution of this paper is in the development of vision-based navigation and integrated control system for straight-running or turning behavior to guarantee performance during of the working. The presented system benefits from a magnetic compass and a fixed camera. An estimation method of boundary of the workspace distance obtains view of in-front-of the robot using a monocular camera to enable detect edges of the workspace. Moreover, an obstacle detection method obtains in-front-of view of the robot using the camera. In this method is applied a deep-learning module to detect and cognition an obstacle. A rotation controller is developed to counter-rotation turn the robot such that at the target boundary point, to avoid the robot reach out of the workspace, or avoid an obstacle. In verification experiment, a state-action space, including the position of the obstacle was constructed and can avoid an obstacle, in each work. Further, we have been confirmed that the robot can counterrotation turn 90 degrees (spin-turn) at the boundary point of workspace.