Tightly-coupled Data Fusion of VINS and Odometer Based on Wheel Slip Estimation

Abstract

The data fusion of a monocular visual-inertial system (VINS) and encoder measurements has proved to be significantly effective in overcoming the additional unobserv-ability of scale, when the robot is constrained to move with constant acceleration on the ground. However, the encoder measurements for positioning may become unreliable once the ground vehicle exhibits wheel slippage. As a result, extending VINS to incorporate such faulty odometer measurements directly could lead to a deterioration of the localization performance. To address this issue, we firstly present a wheeled mobile robot model that relaxes the pure rolling hypothesis for slip estimation. We then propose an adaptive strategy based on the slip estimation to combine acceptable encoder measurements with VINS. Experimental results are presented that demonstrate the reliable estimation of the wheel slip, as well as the improvement of the proposed data fusion scheme in positioning performance.