机器人学研究室知识库

高精度实时状态估计是无人机安全飞行及执行各种任务的首要条件.多传感器(如视觉、IMU和GPS等)融合可提高状态估计精度,并实现信息冗余,当其中某些传感器出现故障时,仍具有较好的鲁棒性.因此,本文提出结合滤波与优化的无人机多传感器融合方法,从而得到局部高精度、全局无漂移的状态估计.该方法主要分为卡尔曼(Kalman)滤波和全局优化两部分.卡尔曼滤波器作为主体融合框架,融合局部传感器(inertial measurement unit, IMU)和全局传感器(经优化后的视觉、GPS、磁力计和气压计)信息得到全局位姿估计.由于卡尔曼滤波算法计算量较小,可以保证融合估计的实时性.全局优化则负责将有漂移的视惯里程计(视觉惯性里程计)信息与全局传感器(GPS,磁力计和气压计)融合对齐后,得到高精度的全局视觉估计.但优化输出会出现不连续且视觉处理存在延迟的问题.因此,将优化后的里程计再输入到卡尔曼滤波器中,从而得到高精度、实时无漂移的状态估计.最后结合具体无人机平台,进行了实际的飞行测试与定位实验,实验结果表明了本文方法的有效性和鲁棒性.

Accurate and real-time state estimation is the first step for UAV (unmanned aerial vehicle) safe flight and operation. Multiple sensors fusion, such as vision, IMU and GPS, can improve state estimation accuracy and can still provide output even some sensors becoming fault or unavailable. Thus, this paper proposes a multiple sensors fusion method based on the combination of filter and optimization to achieve locally accurate and globally drift-free state estimation. The method is divided into two main parts: Kalman filter and global optimization. Kalman filter is considered as the main structure of fusion framework, which fusing local sensor (IMU) and global sensors (aligned global visual inertial odometry, GPS, magnetometer and barometer) to obtain real-time global state estimation. Global optimization is to estimate the transformation between local base frame of visual inertial odometry and global base frame to obtain accurate global visual estimation. However, given the discontinuity of optimization and the delay of odometry, the aligned visual odometry is input into Kalman filter to achieve accurate, real-time and drift-free state estimation. Finally, flight and localization tests on practial UAV are conducted, and the experimental results demonstrate the effectiveness and robustness of the proposed multiple sensors fusion method.