Abstract:

To enhance the efficiency of scene-based test as a safety verification means of automated driving system， a parallel-computing-based accelerated test method for automated driving system is proposed in this paper. Firstly， a parallel accelerated test framework is constructed consisting of top management layer， middle coordination layer and bottom execution layer. Then the specific tasks of different layers are determined： including the middle-tier coordination coverage and the specific test parameters of bottom-tier execution； top layer for the management of unit flow； middle layer for the coordination of parameter calculation； while the bottom layer for the execution of testing. Finally， with a scene of front vehicle cutting-in as an example， a comparative testing with four modes， i.e. parallel accelerated， parallel traversal， single line accelerated and single line traversal is conducted on a black-box automated driving algorithm. The results show that in a condition of totally 4 590 testing scenes， all dangerous scenes are found by four modes， but with significant difference in consumed time， being 1.3， 5.7， 22.4 and 96.2 h respectively， demonstrating that the parallel accelerated testing method proposed can effectively and rapidly discover hazardous scenes in parameter space， enhancing the testing efficiency of automated driving system.

Key words: vehicles, automated driving system, accelerate testing, parallel testing, three layers accelerate testing framework