Improved Multi-IMU-to-Vehicle Extrinsic Calibration Based on a Rigid-Body Covariance Model

Autonomous vehicles rely on a set of sensors to provide critical information about their state and about the state of the surrounding environment. In order to provide useful information, these sensors first need to be properly calibrated both intrinsically and extrinsically. Therefore, extrinsic calibration is essential to safe and effective navigation of an autonomous vehicle. However, manual calibration is an expensive and time-consuming process, which limits the autonomy of the platform. Prior work has developed methods of online, autonomous extrinsic calibration for a single IMU, and for networks of IMUs. Even the state-of-the-art methods require either an unrealistic number of IMUs to be present in the sensor suite, or a calibration maneuver which is difficult to perform in a typical urban environment where it might be deployed. This paper introduces an improved method of Multi-IMU sensor-to-vehicle extrinsic calibration. A three stage weighted least squares algorithm is used to fuse information from a network of IMUs, generating an improved combined sensor pose estimate. This method has been validated through a battery of tests in Carsim 9, a high fidelity vehicle dynamics simulation engine. The performance of the calibration algorithm has been further characterized as a function of the number of IMUs in the network, and the magnitude of the measurement noise. The results from these tests demonstrate an improvement in the extrinsic calibration accuracy over existing methods, even when the calibration routine is limited to a single dynamic maneuver.