GNSS/INS Integration Based on Machine Learning LightGBM Model for Vehicle Navigation

To solve the problem of data accuracy degradation vehicle GNSS/INS integrated navigation systems when GNSS signal is unavailable or there a outage, this paper improves existing integration methodology for land based on AI method. First, position update architecture (PUA) using LightGBM regression predicting during outage presented. It uses to model relationship between INS and changes. On-board are collected available used train PUA-LightGBM model; in event as input predict change position. Second, acquisition system was designed validation. This included self-developed Novatel pwrpak7-e1 six road segments. Finally, were machine learning training PUA-RandomForest model. As result, predicts with less error takes time train. also demonstrated that by allowing be dynamically trained updated while moving could adapt perfectly predictions changes different complex