RIDI: Robust IMU Double Integration (1712.09004v2)

Abstract: This paper proposes a novel data-driven approach for inertial navigation, which learns to estimate trajectories of natural human motions just from an inertial measurement unit (IMU) in every smartphone. The key observation is that human motions are repetitive and consist of a few major modes (e.g., standing, walking, or turning). Our algorithm regresses a velocity vector from the history of linear accelerations and angular velocities, then corrects low-frequency bias in the linear accelerations, which are integrated twice to estimate positions. We have acquired training data with ground-truth motions across multiple human subjects and multiple phone placements (e.g., in a bag or a hand). The qualitatively and quantitatively evaluations have demonstrated that our algorithm has surprisingly shown comparable results to full Visual Inertial navigation. To our knowledge, this paper is the first to integrate sophisticated machine learning techniques with inertial navigation, potentially opening up a new line of research in the domain of data-driven inertial navigation. We will publicly share our code and data to facilitate further research.

Robust IMU Double Integration (RIDI): An Essay on a Novel Approach to Inertial Navigation

The paper introduces a data-driven approach to inertial navigation that leverages Inertial Measurement Units (IMUs) found ubiquitously in smartphones. Named Robust IMU Double Integration (RIDI), this method aims to tackle the longstanding challenge presented by sensor errors and biases associated with traditional inertial navigation systems, which severely compromise the accuracy of motion estimation.

Core Contributions

RIDI distinguishes itself by integrating machine learning techniques to infer a velocity vector from historical IMU data. The system operates in two stages:

1. Velocity Regression: It employs a Support Vector Machine (SVM) to classify smartphone placement types, such as in a pocket or a bag, and then uses Support Vector Regression (SVR) models to estimate velocities across the classified placements.
2. Acceleration Correction: It models errors in linear accelerations as low-frequency corrections, which are calculated using linear least squares to align the velocities derived from integration to those obtained from regression.

Contrary to conventional methods relying solely on raw IMU data, RIDI incorporates predictive learning to rectify errors iteratively, thus reducing positional drift.

Evaluation and Results

The RIDI system shows positional error rates below 3%, considerably outperforming other baseline methods, such as raw double integration and step-counting techniques. This notable accuracy confirms RIDI's potential equivalency to visual-inertial odometry systems while maintaining energy efficiency and operational readiness in conditions where visual methods falter, such as low light environments.

RIDI's success is epitomized by accurate trajectory estimation over extensive sequences, even under varied placements and walking styles. Furthermore, the introduction of a robust, data-driven approach enables generalization across different devices and subjects, indicating flexibility beyond the constraints of specific device or motion patterns.

Implications and Future Work

The implications of RIDI are multifaceted:

• Commercial Applications: Given the pervasive presence of IMUs in consumer electronics, the RIDI approach holds promise for enhancing location-based services without demanding additional energy consumption, thus enabling persistent and accurate navigation solutions.
• Research Continuations: RIDI paves the way for further exploration in data-driven inertial navigation, offering potential frameworks for integrating deeper machine learning methodologies.
• Performance Optimization: Future work involves enhancing generalization capabilities through broader data collection and refining regression models to cover a diverse range of devices and user behaviors. Additionally, porting RIDI to run efficiently on mobile devices is anticipated to broaden its practical applicability.

Conclusion

Through its innovative framework and successful evaluations, RIDI emerges as a promising direction for handling the complex challenges of inertial navigation in smartphones. While ongoing research will aim to overcome current limitations, this paper sets a substantial foundation for future developments in motion estimation technologies, inviting a paradigm shift towards machine learning-driven solutions in inertial navigation.