Collaborative Multirotor UAVs Subject to Disturbances for Precise Maneuvers

航空工业正在经历重大变化，对无人机（UAV）的兴趣增加，这正在创造重大的积极社会和经济影响。加拿大交通部2025年无人机战略的五个关键优先事项之一是制定战略，通过支持加拿大无人机行业的发展，使尖端（遥控）无人机技术为国际市场做好准备。市场上的下一代无人机是仍处于研究阶段的自主无人机。自主无人机没有人类飞行员的智能，必须有一个强大而可靠的系统来执行关键任务，特别是在具有挑战性的拥挤城市环境或在不同阵风和通信错误下的关键军事任务。尽管取得了进展，但在真实的场景中，还没有一个强大的系统来大规模部署自主无人机。 该研究计划将有助于协作，高度可重复的多旋翼无人机的建模，估计，控制和路径规划，用于自主探索受到各种干扰的未知环境。外部干扰，包括邻近多旋翼无人机的影响;在墙壁、天花板或地面附近飞行;以及阵风，都可能严重降低多旋翼无人机的性能，从而导致完全故障。目前的空气动力学模型不能在机载计算模块上实时预测多旋翼无人机不同可能构型的空气动力学效应。因此，在该计划中，人工智能（AI）和机器学习（ML）的进步将用于增强物理模型，以涵盖复杂的未知物理，特别是空气动力学相互作用。然而，目前的AI/ML技术并不完全适用于航空航天行业，它们需要改进和调整，因为这个行业不能容忍任何风险。 人工神经网络（ANN）是AI/ML最强大的工具之一。虽然基于人工神经网络的算法有巨大的进步，但在非欧几里德流形上没有合适的人工神经网络算法来训练高度可扩展的无人机而不简化其动力学。该计划致力于为无人机开发人工神经网络，使其能够执行需要大而快的旋转的精确机动;这考虑了无人机的真实的动力学，并导致更精确的运动规划和轨迹跟踪。人工神经网络将被开发用于非欧几里得空间，以增强无人机的基于物理的模型，使其能够自主地遵循受各种外部干扰的期望路径，并在混乱的环境中生成更有效的路径。非欧几里得空间的人工神经网络将有利于水下机器人控制，几何计算机视觉，计算机动画和人形机器人，以及。

The Aeronautics industry is undergoing significant changes with an increase in the interest in unmanned aerial vehicles (UAVs), which is creating significant positive social and economic effects. One of the five key priorities of Transport Canada's Drone Strategy for 2025 is to develop strategies to enable cutting-edge (remotely piloted) drone technologies to prepare for international markets by supporting the growth of the drone sector in Canada [11]. The next generations of UAVs for the market are Autonomous UAVs that are still in the research phase. An autonomous UAV, without the intelligence of a human pilot, must have a robust and trustworthy system to perform critical missions, especially in challenging crowded urban environments or critical military missions under different wind gusts and communication errors. Despite developments, there is no robust system for mass deployment of autonomous UAVs in real scenarios. This research program will contribute to the modelling, estimation, control and path planning of collaborative, highly maneuverable multi-rotor UAVs for autonomous exploration of unknown environments subject to a variety of disturbances. External disturbances, including effects of adjacent multi-rotor UAVs; flying near walls, ceilings, or the ground; and wind gusts can deteriorate the performance of multi-rotor UAVs substantially, which may result in complete failure. Current aerodynamics models cannot predict aerodynamics effects for different possible configurations of multi-rotor UAVs in real-time on onboard computing modules. Thus, in this program, the advances in artificial intelligence (AI) and machine learning (ML) will be used to enhance the physics models to cover the complex unknown physics, especially aerodynamics interactions. However, current AI/ML techniques are not perfectly suitable for the aerospace industry, and they need to be improved and adjusted since no risk is tolerated in this industry. Artificial Neural Networks (ANN) are one of the most powerful tools of AI/ML. While there are huge advances in algorithms based on ANN, there is no appropriate ANN algorithm on the non-Euclidean manifold to train highly maneuverable UAVs without simplifying their dynamics. This program is committed to the development of ANN for UAVs able to perform precise maneuvers that require large and fast rotations; this takes into account the real dynamics of UAVs and results in more accurate motion planning and trajectory tracking. The ANN will be developed for non-Eucleadian spaces to augment the physics-based model of the UAV to enable it to autonomously follow the desired path subject to various external disturbances and to generate more efficient paths through cluttered environments. The developed ANN for non-Eucleadian spaces will benefit underwater vehicle control, geometric computer vision, computer animation and humanoid robots, as well.