AI-IO: An Aerodynamics-Inspired Real-Time Inertial Odometry for Quadrotors

TL;DR

Inspired by the aerodynamics model and IMU measurement model, the key physical quantity--rotor speed measurements required for inertial odometry is identified and a transformer-based inertial odometry is designed and validated across multiple datasets and real-time systems.

Abstract

Inertial Odometry (IO) has gained attention in quadrotor applications due to its sole reliance on inertial measurement units (IMUs), attributed to its lightweight design, low cost, and robust performance across diverse environments. However, most existing learning-based inertial odometry systems for quadrotors either use only IMU data or include additional dynamics-related inputs such as thrust, but still lack a principled formulation of the underlying physical model to be learned. This lack of interpretability hampers the model's ability to generalize and often limits its accuracy. In this work, we approach the inertial odometry learning problem from a different perspective. Inspired by the aerodynamics model and IMU measurement model, we identify the key physical quantity--rotor speed measurements required for inertial odometry and design a transformer-based inertial odometry. By incorporating rotor speed measurements, the proposed model improves velocity prediction accuracy by 36.9%. Furthermore, the transformer architecture more effectively exploits temporal dependencies for denoising and aerodynamics modeling, yielding an additional 22.4% accuracy gain over previous results. To support evaluation, we also provide a real-world quadrotor flight dataset capturing IMU measurements and rotor speed for high-speed motion. Finally, combined with an uncertainty-aware extended Kalman filter (EKF), our framework is validated across multiple datasets and real-time systems, demonstrating superior accuracy, generalization, and real-time performance. We share the code and data to promote further research (https://github.com/SJTU-ViSYS-team/AI-IO).

Figures (9)

• Figure 1: Comparison of our realtime AI-IO with Intel T265's visual-inertial odometry (VIO) and AirIO. Our method achieves comparable accuracy to VIO under normal lighting and superior performance under dark conditions, demonstrating stronger robustness to visual degradation, while also achieving higher accuracy than state-of-the-art AirIO.
• Figure 2: Proposed transformer-based architecture: Verified measurements are normalized, encoded, and fed into a transformer to capture spatiotemporal dependencies. Finally, two fully connected decoders predict velocity and uncertainty.
• Figure 3: Hardware setup which assembles a low-cost BMI270 IMU for acceleration and angular velocity and ESCs using bidirectional Dshot for rotor speed.
• Figure 4: (a) Ablation results for each module on the test dataset. (b) Comparison of online and offline inference results between AI-IO and AirIO with a window length set to 1 second.
• Figure 5: Comparison of estimated velocity and trajectory in a manual high sequence, with the maximum speed exceeding 10m/s. The ATEs of AI-IO and AirIO are 0.860 and 1.762 respectively, and AI-IO outperforms AirIO by 51.2% in ATE.