Multi-sensor state estimation for autonomous navigation of micro aerial vehicles in GNSS reception deprived environments

Navigation is a key capability of Micro Aerial Vehicles (MAVs). It includes perception, localization, motion control, cognition and obstacle avoidance as the main competences. It may be accomplished by an external operator or onboard flight management system, known as remotely piloted and autonomous systems respectively. The agility of MAVs and complexity of their operating environments have favoured autonomous navigation solutions, which are predominantly Global Navigation Satellite System (GNSS)-based, over remotely piloted solutions. But GNSS technology is susceptible to intentional and unintentional interferences, which challenges have motivated the quest for GNSS-independent autonomous navigation solutions.
Although navigation is supported by several competences, this research focuses on cognition, specifically, the problem-solving intellectual function. Within the navigation task, one of the main problem solvers is the path planner. The key requirement of any path planner is the ability to find feasible paths. Additionally, for MAVs with the inability to conserve power while searching for a path online, the planner ought to be fast and scale well with the environment. This led to the first goal of developing an online path planner with such performance. Another problem to solve arises when the planned path length exceeds the MAV’s endurance, a case common in coverage tasks. For this, a coverage path planner capable of accounting for vehicle endurance in relation to path length and environment size is required. Lastly, autonomous functioning has enabled deployment of mobile robots on our world and beyond. But knowing the right amount of autonomy required to complete a given task is still a challenge. Several autonomy evaluation frameworks have been proposed over the last three decades, but most of these offer a low resolution categorical output or have inconsistent metrics, raising the need for a better autonomy framework.
A path planning ensemble consisting of three concurrently executed single query randomized sampling-based path planners has been proposed for online path planning. A partitioning path planner capable of exact cellular decomposition of large areas of interest into manageable cells using Voronoi decomposition, planning coverage paths and scheduling them on a MAV or a fleet of either homogeneous or heterogeneous MAVs has been proposed as well. Last but not least, a set of four autonomy evaluation metrics, namely capabilities, trust factor, performance capacity and environmental complexity, and their associated mathematical models have also been proposed.
Tested in a physics supported graphical simulator, the proposed path planning ensemble demonstrated query adaptability, a high path finding success rate and a short path planning time, suitable for online path replanning with allowance for path smoothing. Also, the lack of implicit environment representation by sampling-based planners meant that the ensemble planner scales well with the environment. The plausibility of the proposed large-scale coverage path planner has been ascertained through a Software-In-the-Loop (SIL) test. Such a planner guarantees coverage, ensures proper resource management and proper mission planning. The autonomy evaluation framework has been tested on three case studies, which together ascertained the plausibility of its models. This framework provides a systematic approach for development and regulation of autonomy.