Self-driving cars and autonomous driving researchhas been receiving considerable attention as major promisingprospects in modern artificial intelligence applications. Accordingto the evolution of advanced driver assistance system (ADAS),the design of self-driving vehicle and autonomous driving systemsbecomes complicated and safety-critical. In general, the intelligentsystem simultaneously and efficiently activates ADAS functions. Therefore, it is essential to consider reliable ADAS functioncoordination to control the driving system, safely. In order todeal with this issue, this paper proposes a randomized adversarialimitation learning (RAIL) algorithm. The RAIL is a novelderivative-free imitation learning method for autonomous drivingwith various ADAS functions coordination; and thus it imitatesthe operation of decision maker that controls autonomous drivingwith various ADAS functions. The proposed method is able totrain the decision maker that deals with the LIDAR data andcontrols the autonomous driving in multi-lane complex highwayenvironments. The simulation-based evaluation verifies that theproposed method achieves desired performance.