微型无人机（MAV）自主飞行的目标在于实现在复杂环境和复杂任务下的自主导航、规划与控制，并完全依赖机载传感器和计算资源自主完成任务目标。提高MAV在任务全过程中的自主感知、决策与控制等各方面的自主飞行能力，是保证MAV在复杂任务要求与任务场景下满足应用需求的关键。由于复杂封闭环境的特性以及MAV自身的约束，该类环境中的MAV自主飞行问题存在诸多特殊性，对MAV的自主能力提出了诸多新的挑战。本文针对复杂封闭环境下的MAV自主飞行问题，围绕自主轨迹规划、自主导航和状态估计、MAV系统设计构建三个基本方面进行了系统研究。本文的主要研究工作和创新点如下： 一、针对非结构化复杂三维环境下的MAV轨迹规划问题，本文提出了一种基于混合采样的随机轨迹规划算法（RSB-MRRT）。该算法基于一种混合采样策略对经典RRT架构进行了改进：该策略通过RSB策略辨识环境中的狭窄通道区域，以提高狭窄通道结构内的采样点密度，同时结合均匀采样的优势以产生更为合理的采样点全局分布；在此基础上结合多RRT架构设计了轨迹规划算法，从而提高了算法在复杂非结构化环境下生成连通路径的效率。 二、针对不确定环境下存在状态估计不确定性的MAV轨迹规划问题，本文设计了一种改进的MAV实时轨迹规划方法(CL-RBT)，一方面继承了基于闭环系统模型的状态预测策略以及实时规划-执行架构，以有效应对环境的不确定性，并可兼顾MAV的动力学约束；同时在规划过程中引入了状态估计不确定度的衡量，使生成的轨迹在降低路径代价的同时可有效约束MAV的定位和状态估计不确定性。 三、本文设计了一种基于RGB-D相机和惯性器件的MAV组合导航框架。首先，本文采用了一种改进的RGB-D鲁棒相对运动估计算法，在特征检测、特征匹配和相对运动估计方面对现有算法进行了改进，提高了相对运动估计的鲁棒性和算法执行效率；基于不变观测器理论设计了RGB-D/惯性组合导航状态估计算法，实现了RGB-D运动状态估计值与惯性测量参数的融合，并可在无GPS外部辅助导航源的情况下，单纯依靠机载设备和计算资源实现MAV的自主导航状态估计。 四、设计了一种高自主性、低成本、开放式的MAV自主飞行系统框架，并搭建了四旋翼MAV系统。飞行实验证明该MAV系统可在无GPS等外部辅助导航源的室内环境中，依靠机载传感器和计算资源实现自主飞行。另一方面该系统为进一步研究和验证MAV自主飞行领域不同层级的相关理论和方法提供了基础。

The objective of autonomous flight of micro aerial vehicles (MAVs) is to achieve autonomous navigation, planning, guidance and control that rely solely on onboard sensing and computational capabilities. In particular, autonomous trajectory planning and navigation are key enabling technologies for the autonomous flight of MAVs. Due to the special constraints imposed by enclosed complex environments and the characteristics of MAVs, the autonomous flight of MAVs still faces various unique challenges. In order to enhance the autonomous flight capabilities of MAVs in uncertain, dynamic and complex environments, this dissertation investigates the autonomous trajectory planning, state estimation as well as MAV system design on the basis of analyzing limitations of previous research and existing technologies. The primary research and contributions of this thesis are summarized as follows:1. To address the trajectory planning of MAVs in unstructured complex environments, a randomized trajectory planning approach (RSB-MRRT) is developed based on a hybrid sampling strategy (RSB-Uniform). Taking advantage of the RSB test algorithm uniform sampling, the RSB-Uniform is able to increase the sampling density in narrow passage structures while generating a more desirable global distribution of milestones. In addition, a multiple RRT-based trajectory planning framework is employed to expand and connect multiple trajectories. Simulation results demonstrate that the proposed RSB-MRRT outperforms other existing approaches in terms of the overall efficiency of trajectory planning in unstructured environments with narrow passages.2. A novel sampling based trajectory planning approach (CL-RBT) is proposed for MAVs in GPS-denied, uncertain environments. The CL-RBT extends the conventional closed loop RRT algorithm by employing sensor characteristics and a qualification of state estimation uncertainty, which is predicted using a linear covariance propagation approach. As a result, the CL-RRT is able to generate dynamic feasible trajectories that trade off between minimizing trajectory cost and reducing state estimation uncertainty. In addition, the CL-RBT inherits the real time planning-execution scheme of conventional closed loop RRT, enabling the trajectory planning algorithm to account for MAV dynamics, as well as dynamic and uncertain environments3. A RGB-D/inertial navigation framework is proposed for state estimation of MAVs in GPS-denied environments. A robust RGB-D relative motion estimation approach is developed to improve the performance of existing algorithms in terms of robustness and efficiency in feature extractions, feature matching and relative motion calculation. Additionally, the above RGB-D motion estimates are fused with IMU measurements using the invariant observer scheme, yielding a full estimation of the MAV’s attitude, position and velocity. Indoor flight test results demonstrate that the proposed RGB-D/inertial navigation framework is able to provide reliable and accurate state estimates for MAVs, without relying on any external navigation aids.4. A low-cost, scalable MAV system framework is designed and a quadrotor-based actual MAV system is developed accordingly. Based on the quadrotor MAV system, a number of flight experiments are conducted to validate the effectiveness of the proposed trajectory planning and navigation approaches. This MAV system integrates the aforementioned design/technologies and provides a scalable framework for further investigation and evaluation of relevant approaches and technologies for autonomous MAVs.