Autonomous Indoor Timber Stocktaking with Multi-rotor Unmanned Aerial Vehicles(UAV)

Abstract

A timber warehouse is considered to be a hazardous environment due to handling and storage of heavy wood products. Stocktaking in such an environment is risky to human lives due to the involvement of manual scanning of barcodes placed on products stored at high altitudes. A solution to this safety issue is to use multi-rotor Unmanned Aerial Vehicles (UAVs) for stocktaking. However, current methods are found to be semi-autonomous with no awareness of the surroundings, which is a significant disadvantage for UAVs operating in dynamic environments such as an indoor timber warehouse. Furthermore, other methods use a pre-scanned map of the environment. This thesis describes the development of a Vision Guidance Controller for autonomous navigation inside a GPS-denied indoor timber warehouse using online path planning. An in-depth analysis of an indoor timber warehouse helped in the identification of critical observable features of the surroundings. The obtained environment information is used to select the complementary perception sensors of the UAV, which enable environment awareness. Perception methods utilised in this research are primarily focused on the detection of vertical surfaces, horizontal and physical edges present in the environment. Since the UAV requires to scan barcodes present on a finite vertical non-flat surface, in this work, the stocktaking process is treated as a coverage path planning problem. The primary outcomes of this work include the implementation of machine vision techniques for the detection of environmental features and the development of a vision-based control method to guide the UAV for successful coverage of a vertical non-flat region of interest. A non-decomposition based coverage path plan is deployed by the controller which utilises a geometric pattern to plan the UAV's path for complete area coverage dynamically. The complete Vision Guidance Controller is implemented using a Moore Finite State Machine (MFSM). Performance of the developed controller is quantified through indoor experimental testing with the help of an in-house fabricated surrogate environment whose features are varied to match the realworld setting. The proposed system is found to be highly effective at stocktaking with a coverage performance of 100% for the tested flight velocities of up to 0.5 ms⁻¹.