Vision-based high-speed manipulation for robotic ultra-precise weed control

In this system paper we present a novel approach for autonomous, mobile manipulation for agricultural robots. Our target application is mechanical weed control which for example is needed in organic farming. Today, this task is often performed by field workers, whose availability is declining and the quality of their work differs greatly. In addition, manual weeding is a very tedious task with adverse health effects. Hence, mechanical weed control is considered as a prime example for automation. One of the challenges in order to perform the task efficiently is the need for fast plant detection and precise treatment of single plants with a mobile manipulator. The unstructured environment and rough field conditions raise additional challenges. We introduce a first system concept and control design known to handle these issues. In particular, high speed image processing and visual servoing are applied to precisely position a specially designed weeding tool. To evaluate the performance of the mobile manipulator we run experiments with a mobile test vehicle in field-like conditions and demonstrate that our system can treat single plants in less than 1s as it is required to be competitive.