AI-IMU Dead-Reckoning (1904.06064v1)

Abstract: 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

• The paper introduces an integrated IEKF and deep learning approach that dynamically optimizes IMU data for vehicular dead-reckoning.
• The method leverages pseudo-measurements and adaptive covariance tuning, achieving a competitive 1.10% translational error on the KITTI dataset.
• The approach enhances real-time vehicle localization under sensor failure conditions, reducing reliance on costlier systems like LiDAR and stereo vision.

AI-IMU Dead-Reckoning: A Summary

The paper presents a method for vehicular dead-reckoning using only an Inertial Measurement Unit (IMU), aimed at enhancing the functionality of intelligent vehicles. IMU-based dead-reckoning is crucial for integrating and validating data from imaging sensors and for ensuring navigation through challenging environments, particularly when external sensor systems fail.

Methodology

The core of this approach lies in combining an Invariant Extended Kalman Filter (IEKF) with deep learning techniques to optimize the IMU data interpretation dynamically. The IEKF algorithm updates its estimates of the vehicle's state, including position, velocity, and orientation, by integrating IMU measurements and incorporating pseudo-measurements that account for vehicle dynamics. These pseudo-measurements assume low lateral and upward velocities, typical for ground vehicles moving on leveled paths, but the validity of these assumptions is dependent on the driving context.

Key Components:

• Invariant Extended Kalman Filter (IEKF): Utilized for integrating IMU readings and pseudo-measurements. This variant of the EKF is particularly effective due to its robustness in handling the specific geometrical constraints and noise characteristics of the IMU data.
• Deep Neural Networks (DNN): Deployed to dynamically adjust the noise parameters of the IEKF. This network process sequences of IMU data and modifies the filter's covariance parameters to improve accuracy based on the driving conditions.

Experimental Evaluation

Testing on the KITTI odometry dataset demonstrated that this method achieves a translational error of approximately 1.10%, matching or surpassing the performance of competing methods that use LiDAR or stereo vision. This accomplishment highlights the potential for IMU-only methods to reduce cost and complexity in vehicular positioning systems while maintaining competitive accuracy.

Practical and Theoretical Implications

This methodology provides a valuable tool for real-time localization in autonomous vehicles. It offers an inherent advantage in situations where visual or LiDAR inputs are compromised, enhancing the robustness of localization systems. The adaptability of the Kalman filter's covariance matrix through deep learning is particularly noteworthy, offering a promising avenue for future research in AI-driven sensor calibration.

Future Directions

The research suggests several potential advancements. Future exploration could focus on extending this method to other vehicular types or incorporating global navigation satellite system (GNSS) data for even higher precision. Additionally, refining the neural network architecture to reduce computational demands without sacrificing accuracy could further increase the method's practicality.

In summary, the integration of IEKF and advanced AI techniques in this paper presents a sophisticated approach to vehicular dead-reckoning, showing promise for future developments in localizing intelligent vehicles.