摘要： In most of visual-inertial state estimation frameworks, camera is usually assumed synchronized with inertial measurement unit (IMU). However, in practice, the timestamps of camera and IMU are typically affected by a time-varying delay, especially for low-cost IMU-camera integrated system (e.g. android phone). Tracking failures could occur easily then. In this paper, we discuss the effect of the temporal offset on map points triangulation and nonlinear optimization during visual-inertial simultaneous localization and mapping (SLAM) process, and propose two feasible solutions for compensating the offset between camera and IMU. We choose the simpler solution and propose a computationally tractable approach which can estimate the offset with all other variables of interest (the IMU pose, the camera-IMU calibration, etc.). Our simulation results demonstrate that the proposed approach is effective in scenarios involving both constant and time-varying offsets.