This repository provides an AI-IMU Dead-Reckoning system for wheeled vehicles, targeting researchers and engineers in autonomous navigation. It achieves low translational error on the KITTI dataset using only an IMU, offering a robust alternative to LiDAR or stereo vision systems, especially during sensor failures or obstructions.

How It Works

The system employs a Kalman filter integrated with a deep neural network. The Kalman filter fuses IMU measurements with zero lateral and vertical velocity constraints to refine state estimates. The novel aspect is the deep learning adapter, which dynamically adjusts the filter's noise parameters by directly mapping raw IMU signals to covariance matrices, eliminating the need for explicit state estimates.

Quick Start & Requirements

• Install with pip3 install matplotlib numpy termcolor scipy navpy.
• Requires Python 3.5+.
• Download KITTI IMU data and training parameters from provided URLs.
• Run tests with cd ai-imu-dr/src && python3 main_kitti.py.

Highlighted Details

• Achieves 1.10% translational error on KITTI odometry sequences using only an IMU.
• Self-calibrates IMU biases.
• Competes with LiDAR/stereo vision-based methods.
• Deep learning adapter adapts noise parameters in real-time.

Maintenance & Community

The project is authored by Martin Brossard, Axel Barrau, and Silvère Bonnabel from MINES ParisTech and Safran Tech. A related IEEE paper is available.

Licensing & Compatibility

The repository does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The code was tested under Python 3.5, which is now end-of-life. The project's maintenance status and community activity are not detailed in the README.