AI-IMU Dead-Reckoning
About
In this paper we propose a novel accurate method for dead-reckoning of wheeled vehicles based only on an Inertial Measurement Unit (IMU). In the context of intelligent vehicles, robust and accurate dead-reckoning based on the IMU may prove useful to correlate feeds from imaging sensors, to safely navigate through obstructions, or for safe emergency stops in the extreme case of exteroceptive sensors failure. The key components of the method are the Kalman filter and the use of deep neural networks to dynamically adapt the noise parameters of the filter. The method is tested on the KITTI odometry dataset, and our dead-reckoning inertial method based only on the IMU accurately estimates 3D position, velocity, orientation of the vehicle and self-calibrates the IMU biases. We achieve on average a 1.10% translational error and the algorithm competes with top-ranked methods which, by contrast, use LiDAR or stereo vision. We make our implementation open-source at: https://github.com/mbrossar/ai-imu-dr
Key insights
How does AI-IMU Dead-Reckoning estimate vehicle position using only an IMU?
It tracks a vehicle's position, velocity, and orientation using only an inertial sensor. A Kalman filter combines motion physics with driving constraints, while a neural network dynamically adjusts filter noise parameters to adapt to maneuvers like turning.
How does this IMU-only method compare to LiDAR and stereo vision systems?
Unlike prior methods needing LiDAR or cameras, this uses only a moderate-precision IMU. On the KITTI dataset, it achieved 1.10% average translational error, outperforming ORB-SLAM2 (1.17%) and approaching the LiDAR-based IMLS (0.98%).
What practical applications does this dead-reckoning approach support?
The authors motivate backup motion estimation when cameras or other sensors fail, and discuss railway vehicles, GNSS fusion and map matching. Experiments concern vehicle odometry; they do not establish the safety of a deployed emergency-stop system.
What limitations and future work do the authors identify?
Process noise parameters are fixed rather than dynamically adapted. Sequences with missing data showed degraded performance. Future work includes learning Kalman covariance for images and addressing generalization across different vehicles.