Evaluation of multisensor odometry and SLAM free of ground truth

The evaluation of odometry and Simultaneous Localization and Mapping (SLAM) systems traditionally relies on the availability of highly accurate ground-truth (GT) trajectories, which can be difficult or impossible to obtain in large-scale or challenging environments. To address this limitation, this paper introduces a novel, GT-free framework for evaluating multi-sensor based odometry and SLAM systems by analyzing the consistency between estimation results generated by different sensor sources. We propose two metrics for this purpose: Frame Jerk, which quantifies local pose estimation volatility, and Sliding-Window Drift, which measures the accumulation of drift over variable trajectory segments. The proposed metrics are versatile, applicable to both loosely-coupled and tightly-coupled (both filtering-based and graph optimization-based methods) estimation paradigms. Experimental results demonstrate that our consistency metrics provide a reliable measure effectively correlating with GT-based measures or navigation errors.