Linear Likelihood Approach-Based Adaptive Fusion of Heterogeneous Multiple IMUs for Orientation Estimation

Multiple inertial measurement unit (multi-IMU) fusion plays a crucial role in various applications requiring accurate motion estimation. This work presents a likelihood-based approach for the fusion of multiple heterogeneous inertial measurement units (IMUs) that adapt based on residual computed by individual complementary filters (CFs). This method utilizes a residual-based weighting architecture (RWA) to adjust the weights for motion estimates from different IMUs dynamically in response to varying conditions. The axis-level architecture proposed here ensures different weights for orientation estimates from different axes in each IMU. The proposed method is evaluated using a simulation setup for five different scenarios that are, random noise, bias, sudden drift, oscillation, and abrupt freezing conditions. These conditions mirror real-world challenges faced by inertial sensors during motion. It has also been tested with different datasets and benchmarked with the traditional averaging-based scheme for multiinertial sensor fusion. The experimental analysis depicts a noticeable enhancement in accuracy, as evidenced by the reduction in root mean square error (RMSE) values achieved through the proposed method.