The efficient detection of the performance limit is critical to automatic driving systems. With the motivation that automatic driving is more difficult to be realized under more complicated scenarios, an improved genetic algorithm (IGA) based evolution test is proposed to accelerate the test process for the performance limit evaluation of automatic driving systems. IGA conducts the crossover operation at all positions and the mutation operation for several times to make the high quality chromosome exist in the candidate offspring easily. Then the normal offspring is selected statistically based on the index of scenario complexity, which is designed to measure the difficulty of automatic driving indirectly by using the Analytic Hierarchy Process. The benefits of the modified cross/mutation operators on the improvement of the scenario complexity are analyzed theoretically. The effectiveness of IGA based evolution test is validated by application to the evaluation of the collision avoidance performance of an automatic parallel parking system. The simulation results show that compared with the traditional genetic algorithm, the average scenario complexity of the offspring is increased by 16.59%, the convergence speed is about twice as fast and IGA can find the collision condition successfully.