Positioning Algorithm with Tightly Coupled Wheel Odometry and VIO in Off-Road Environments

针对视觉惯性里程计(Visual-Inertial Odometry, VIO)在越野环境中定位性能显著下降的问题，本文提出了一种基于轮速里程计与VIO紧耦合的算法HW-VIO (Hybrid Wheel-VIO)。该算法融合了IMU与轮速里程计的特点，设计了混合预积分观测模型，并利用轮速里程计的零速更新校正IMU加速度计和陀螺仪的偏置误差。为改善轮速计异常值频发的问题，本文引入卡方检验算法，对混合预积分残差进行评估，从而稳健识别并剔除异常数据。最后，在三种难度不同的野外农田场景中对算法进行了测试。实验结果表明，本文算法能够显著提高VIO系统的性能，平均定位精度提升47%。此外，通过消融实验进一步验证了混合预积分观测模型的有效性，相较于直接进行轮速融合的W-VIO (Wheel-VIO)算法，平均定位精度提升达50%。The declining localization performance of Visual-Inertial Odometry (VIO) in off-road environments is a significant challenge. To address this issue, a tightly coupled algorithm named HW-VIO (Hybrid Wheel-VIO) is proposed, combining wheel odometry and VIO. The method leverages the complementary properties of IMU and wheel odometry by introducing a hybrid pre-integration observation model, where zero-velocity updates from wheel odometry are employed to dynamically correct accelerometer and gyroscope biases in the IMU. To handle the frequent occurrence of outliers in wheel odometry measurements, a chi-squared test is applied to evaluate residuals from the hybrid pre-integration process, enabling robust identification and rejection of abnormal data. The algorithm is validated through experiments conducted in three off-road farmland scenarios with varying levels of difficulty. Results show that HW-VIO significantly improves localization accuracy, achieving an average accuracy improvement of 47%. Furthermore, ablation studies confirm the effectiveness of the hybrid pre-integration model, demonstrating a 50% improvement in localization accuracy compared to the W-VIO (Wheel-VIO) algorithm, which directly fuses wheel odometry.