Autonomous Navigation for Park Maintenance and Greenhouse Horticulture. Robot Driveability, Ultra Wideband Positioning and Calibration, Semantic Navigation

In the conventional manufacturing industry, a new trend in automating traditional processes using modern smart technology has emerged. The new machines of the 21st century allow for a better integration in the manufacturing process, resulting in improved communication and self-monitoring. The whole transition will eventually allow an increase in productivity while keeping the costs under control. When examining the agriculture and horticulture industry and applications for park and garden maintenance, it becomes clear that such transition started to a very limited extent. The automation of these processes requires a focus on the development of mobile solutions for agricultural and outdoor environments. The high variability and unpredictability of such environments are additional challenges. The few available solutions are either very expensive or still require major adaptation to and/or conditioning of the infrastructure. This thesis proposes a general framework for autonomous navigation, which enables the automation of mobile robot applications. The framework is a series of connected components, which support different robot models and allow them to handle various navigation tasks. They also allow to configure different types of environments and integrate a selection of navigation sensors. The content of each component in the framework is a state-of-the-art implementation of a specific algorithm, which can be adapted depending on the considered application. For the benefit of the applications, the platforms are expected to navigate accurately, i.e. follow a desired trajectory with low tracking error. This imposes requirements on the mechatronic design of the system. A performance criterion for the driveability of a differential drive mobile robot is proposed, which allows to validate a conceptual design early in the design process. Applying this performance criterion guarantees a controllable vehicle at first hand, which saves time and money in development. Accurate navigation also requires correct localization of the robot. The applicability of a local positioning system based on ultra wideband (UWB) technology is validated in open areas. The UWB positioning system allows to determine the robot's location accurately with affordable technology. The entire operation of the UWB system is characterized and a novel model for the range bias is defined in order to generate the best possible positioning result. To ensure the ease of use of such a system, a procedure to install and calibrate the UWB system has been developed, which can be performed in a few hours by driving manually a few simple maneuvers. When the robot is deployed in a more structured environment, the localization is solved by defining a map with semantic tags which allow an intelligent programming of feature detection and robot behaviors. The map can be drawn upfront based on an existing floorplan, which allows a fast deployment of a vehicle in a new environment. Due to the variability of outdoor environments, the feature detection itself is solved by extracting structures from a 3D pointcloud. The innovations in this work started from challenges within greenhouse horticulture and park applications, but their applicability is not limited to these cases. The general framework provides a very broad basis that can be used in any other mobile navigation application.

Jaar van publicatie:2021

Toegankelijkheid:Embargoed