DUAL-LIO: Dual-Inertia Aided Lightweight Legged Odometry Using Body Constraints

Low-cost and computationally efficient odometry is essential for enabling robots with limited payload capacity to perform various tasks. Despite many attempts to achieve accurate localization through multisensor fusion or inertial-only methods, achieving a satisfactory balance between odometry accuracy, cost, and computational efficiency remains a significant challenge. In this article, we propose a low-cost, lightweight, and accurate odometry for legged robots, called dual-leg inertial measurement unit (IMU) odometry (DUAL-LIO), which utilizes data from two IMUs mounted on the diagonal legs of the robot. Based on these measurements, the system estimates the attitude, velocity, and position. To further mitigate integration drift, we introduce gait-specific constraints, including zero-velocity constraint and stride constraint derived from the gait patterns. A continuous zero-velocity update (C-ZUPT) method is proposed to seamlessly correct the system throughout the entire process. Furthermore, the stride constraint exploits the maximal leg swing amplitude to improve the accuracy of state estimation. The dual-IMU configuration leverages the inter-IMU distance from their diagonal placement on opposing legs to compensate for odometry errors through spatial constraint. Experimental results validate the effectiveness of our method, demonstrating not only a 46% accuracy improvement but also a significant reduction in computational time compared to state of the art in the public dataset. Our code is available at https://github.com/Demi-xinyi/DUAL-LIO