<strong>Second Place Winner of the Smartphone Decimeter Challenge:</strong> An Open-Source Factor Graph Optimization Package for GNSS and IMU Integration in Smartphones

This paper describes a solution that won second place at the Google smartphone decimeter challenge 2023 (GSDC 2023) held on Kaggle, a machine learning competition platform, from September 13, 2023, to May 24, 2024. GSDC 2023, which has been held annually since 2021, is a competition to estimate a vehicle’s trajectory based on raw global navigation satellite system (GNSS) and inertial measurement unit (IMU) log data from smartphones. In previous competitions, models have been built to estimate smartphone locations by factor graph optimization (FGO) using only GNSS observations. By contrast, in GSDC 2023, we have developed a method for position estimation using tightly coupled GNSS/IMU FGO that adds observations from the smartphone’s IMU. We generated 3D attitude and velocity constraints from IMU acceleration and angular velocity observations and used them for relative position estimation between states. This method is expected to improve position-estimation performance in urban areas where multipath occurs and in environments where positioning accuracy is degraded by GNSS alone, such as elevated structures and tunnels. In addition, by optimizing the pre-processing for each of several smartphone models, the GNSS observation errors based on the smartphone model can be reduced, and the final position estimation accuracy can be improved. After implementing the proposed method and evaluating it in the competition, our final result for GSDC 2023 was 0.789 m private and 0.928 m public, achieving decimeter-accurate smartphone position estimation. The proposed method, which is a combination of tightly coupled GNSS and IMU FGO, is available open source, allowing anyone to reproduce the decimeter-accuracy smartphone positioning method. This will accelerate the realization of high-precision positioning for smartphones.