A Quantitative Assessment Approach for Modeling Complexity of Traffic Scene Tasks

摘要

Constructing a reasonable and effective evaluation system is the key to realize the task complexity from subjective evaluation to objective quantitative analysis. Based on the key common issues of autonomous driving evaluation, this paper uses the definition of dynamic driving tasks, expert evaluation methods, and information entropy theory to design a complex metric for scene-oriented driving tasks from the three aspects of action composition, information perception and judgment and decision. The evaluation method determines the complexity factor of the scene driving task and its quantification method, and establishes the scene task complexity quantification model suitable for seven types of vehicle test verification. The verification results confirm the usability of the model described in this paper.