An autonomous driving vehicle guided safety decision algorithm based on HMM and Gaussian Markov distance

Considering the current actual status of multi-level warning for Autonomous Vehicles (AVs), this paper takes the autonomous driving system of AVs as the research object, and focuses on the whole vehicle state, an AVs Multi-modal Guided Safety Decision (MGSD) algorithm based on Hidden Markov Model (HMM) and Gaussian Mahalanobis Distance (GMD) was proposed. The algorithm defines the safety-related modal of AVs into four levels based on the actual multi-level warning states of AVs. And based on the four-level safety-related modal, Multi-modal definition and Multi-modal Fuzzy Boundaries based on Gaussian Distribution (MFB-G) are proposed. On this basis, Multi-modal Pre-decision Model (MPM) based on GMD and MGSD based on HMM are proposed to realize the automatic guided safety decision for the safety-related modal of AVs. Therefore, the MGSD algorithm proposed does not rely on thresholds and has fewer adjustment parameters and better universality. In addition, it can also consider the individual differences of AVs systems and has a certain degree of online growth potential. Finally, the MGSD algorithm proposed is validated based on mixed abnormal modals of three subsystems in AVs. At the same time, it is compared with a traditional binary pattern detection algorithm based on the residual distribution test. The validation and comparison results indicate that the MGSD algorithm proposed has higher sensitivity, faster response speed, higher detection accuracy, and is more in line with the engineering needs of practical multi-level warning in AVs.