Abstract: Accurate and reliable safety evaluation is the basis for the promotion and application of autonomous vehicles. However, the driving environment is complex and changeable, and traditional methods can’t meet the needs of its safety evaluation. A multi-dimensional logical scenario-oriented clustering evaluation method is established in this paper for autonomous driving safety evaluation. Firstly, the results of ergodic test hazard parameters in a multi-dimensional logic scenario are clustered based on Gaussian model; then, two basic evaluation indicators of the danger field dispersion and danger field range are proposed by using the clustering result. An autonomous driving safety clustering evaluation parameter, namely scenario hazard rate is built by coupling the two indicators. Finally, a black box autopilot algorithm is evaluated using the method proposed in the paper. The results show that the method proposed in this paper can comprehensively consider the statistical law in a multi-dimensional logical scenario, obtain a quantitative index, and achieve a comprehensive scientific evaluation

Key words: autonomous driving, safety evaluation, multi-dimensional logical scenario, cluster analysis, scenario hazard rate