In this work, we propose an interoceptive-only state estimation system for a quadrotor with deep neural network processing, where the quadrotor dynamics is considered as a perceptive supplement of the inertial kinematics. To improve the precision of multi-sensor fusion, we train cascaded networks on real-world quadrotor flight data to learn IMU kinematic properties, quadrotor dynamic characteristics, and motion states of the quadrotor along with their uncertainty information, respectively. This encoded information empowers us to address the issues of IMU bias stability, quadrotor dynamics, and multi-sensor calibration during sensor fusion. The above multi-source information is fused into a two-stage Extended Kalman Filter (EKF) framework for better estimation. Experiments have demonstrated the advantages of our proposed work over several conventional and learning-based methods.

翻译：本文提出了一种仅依赖内感受器的四旋翼状态估计系统，该系统采用深度神经网络处理，将四旋翼动力学作为惯性运动学的感知补充。为提升多传感器融合精度，我们在真实四旋翼飞行数据上训练级联网络，分别学习IMU运动学特性、四旋翼动力学特性、飞行器运动状态及其不确定性信息。这些编码信息使我们能够解决传感器融合过程中的IMU偏置稳定性、四旋翼动力学及多传感器标定问题。上述多源信息被融合到一个两阶段扩展卡尔曼滤波（EKF）框架中，以实现更优的状态估计。实验表明，本文提出的方法在性能上优于多种传统方法和基于学习的方法。