White Paper: Fault detection within autonomous systems
Autonomous vehicles depend on accurate, reliable positioning — but GPS, lidar, and radar all have blind spots. This whitepaper proposes using Inertial Measurement Units as an independent monitoring system to detect when the primary positioning subsystem is failing, enabling fail-safe interventions before they become safety incidents.
Fault detection within autonomous systems
Leveraging Inertial Measurement Units (IMUs) as an independent system to monitor and validate the performance of vehicle positioning systems.
Autonomous vehicles face a critical challenge: ensuring reliable and accurate positioning data to navigate safely in complex environments. Current sensor technologies like GPS, lidar, and radar often struggle to meet the stringent requirements for both safety and continuous availability, particularly under adverse conditions. This whitepaper proposes an innovative fault detection methodology leveraging Inertial Measurement Units (IMUs) as an independent system to monitor and validate the performance of vehicle positioning systems.
The proposed approach uses an Extended Kalman Filter (EKF) to compare motion data from the IMU – such as acceleration and yaw rate – with outputs from the positioning system. By detecting discrepancies and identifying errors, the system enables fail-safe measures like emergency stops, offering a critical layer of safety in automated driving scenarios. This method benefits from the inherent reliability of IMUs, their resistance to external electromagnetic interference, and the potential for redundancy to enhance robustness.
Simulation-based development is employed to test and refine this methodology. Using a dynamic vehicle model, realistic driving scenarios, and fault injection techniques, the paper demonstrates the efficacy of the EKF in detecting certain types of positioning system errors. While preliminary results show promise, identifying some fault types with high accuracy, challenges like sensor bias and drift highlight areas for further refinement.
Tests of the algorithm on a physical platform were possible using a QRT-modified electrical wheelchair and an indoor reference positioning system with cm accuracy, and a commercially available high-performance IMU.
The tested algorithm is a first, novel implementation of the proposed fault detection idea aimed as a first evaluation. As such it only incorporates the estimate of one IMU bias parameter. It is demonstrated both using simulation and a physical system, that this filter has the desired functionality under certain conditions but lacks satisfying performance under other conditions.
Download the full whitepaper for more details about the methodology, test setup and other details.