Learning-based Airflow Inertial Odometry for MAVs using Thermal Anemometers in a GPS and vision denied environment

TL;DR

This work tackles MAV speed estimation in GPS- and vision-denied indoor settings by leveraging airflow sensing with thermal anemometers. A learning-based velocity estimator using a GRU network estimates relative air velocity $V_a^{\{B\}}$ from noisy sensor data, while a neural-network–based acceleration model and a bias-aware observer fuse multi-sensor information (IMU, barometer, ESC) to produce robust odometry. The approach decouples downwash and ground effects and reduces vertical drift, achieving a total position drift of about $5.7$ m over $203$ s in manual flights. The open-source framework demonstrates practical viability for indoor MAV operation without GPS or vision cues.

Abstract

This work demonstrates an airflow inertial based odometry system with multi-sensor data fusion, including thermal anemometer, IMU, ESC, and barometer. This goal is challenging because low-cost IMUs and barometers have significant bias, and anemometer measurements are very susceptible to interference from spinning propellers and ground effects. We employ a GRU-based deep neural network to estimate relative air speed from noisy and disturbed anemometer measurements, and an observer with bias model to fuse the sensor data and thus estimate the state of aerial vehicle. A complete flight data, including takeoff and landing on the ground, shows that the approach is able to decouple the downwash induced wind speed caused by propellers and the ground effect, and accurately estimate the flight speed in a wind-free indoor environment. IMU, and barometer bias are effectively estimated, which significantly reduces the position integration drift, which is only 5.7m for 203s manual random flight. The open source is available on https://github.com/SyRoCo-ISIR/Flight-Speed-Estimation-Airflow.

Figures (7)

• Figure 1: MAV equipped with four thermal anemometers orthogonal to each other.
• Figure 2: System diagram of the proposed approach. The purple blocks are the onboard sensors. The attitude is estimated from the 9-axis IMU. The acceleration estimator, velocity estimator and flight status identifier use similar neural networks, due to their same time-series characteristics. The airflow inertia estimator fuses all available information.
• Figure 3: Neural network architecture of the proposed approach.
• Figure 4: Neural network predictions.
• Figure 5: Data fusion results.