Novel Dual-Kalman-Filter-Based State Estimation Algorithm for Wheel-Legged Multi-mode Autonomous Vehicle

The wheeled-legged multi-mode vehicles (WLMV) possess the capability to dynamically adapt driving modes to various environmental conditions, making them a promising field of research in autonomous vehicles. The high motility of WLMV relies on accurate state estimation, while current research on WLMV state estimation mainly focuses on the single-mode operating scenario. To achieve a comprehensive observation of the WLMV’s states in different modes, a systematic state estimation framework for WLMV has been introduced. Kinematic and dynamic models for WLMVs under different operating scenarios have been formulated. Considering the discrepancy of wheel motion states between wheel mode and leg mode for WLMV, a novel dual Kalman filter estimation (DUKE) algorithm is proposed. In the first layer of DUKE, an interacting multiple model-based Kalman filter (IMM-KF) is designed to modify the wheel motion state. Combining the measured leg motion and estimated wheel motion, the error state Kalman filter (ESKF) is constructed in the second layer to estimate the entire states of the WLMV. A WLMV experimental platform has been constructed to validate the efficacy of the proposed framework across multiple driving modes in different scenarios. Experimental results show that DUKE significantly enhances the accuracy of estimated states compared to traditional single mode state estimation techniques, contributing to the progression of the state-of-the-art in autonomous vehicle technology, especially in the multi-mode transportation system.