Methodology on Comprehensive Mapping of Multi-information of Autonomous Driving Based on Trajectory Tensor

Abstract

The performance of an autonomous driving vehicle is limited by low levels of accuracy and efficiency for information mapping, which can be improved by the optimization of data structure and mapping manner. A trajectory categorical-calculating model for autonomous driving vehicle is built for solving trajectory and attitude by given handling data. Handling information, trajectory information and attitude information are parameterized through multi-fitting. A model of trajectory tensor is proposed according to the characteristics of trajectory parameterization. By taking advantage of the data structure of a tensor with high-order and multi-dimensions, samples of two basic types of mapping in trajectory planning and tracking are worked out. An in-depth learning system is constructed and the mathematic model of the basic relation of two types of mapping is obtained by training offline adequately. In simulation experiments compared with the usual method of differential equations, calculation efficiencies of the trajectory tensor method are higher in all experimental data groups, while the standards required by calculation accuracy are entirely reached. The accuracy and efficiency of trajectory planning and tracking can be improved through the trajectory-tensor-based model of information mapping, and the safety as well as the flexibility of autonomous driving vehicles can be therefore enhanced.