머신러닝을 이용한 자율주행차량의 주행적합지표 설계

초록

In this paper, we classified and composed a set of driving scenarios based on the statistical results of the Traffic Accident Analysis System of the Road Traffic Authority. The driving environmental data and the relative distance between the vehicles were obtained using dSPACE’s ASM (Automotive Simulation Models) and Driving Simulator, respectively. At this time, the relative distance and relative speed were calculated using the position, speed, and acceleration of the autonomous vehicle and the of the surrounding vehicles. Since we need responses of driver or occupant to determine whether autonomous vehicle is appropriate or not, we created an input signal called ‘Trigger Signal’. The driver or occupant trigger it only for inappropriate driving situation of each scenarios. With this experimental data, we built MLP (Multilayer Perceptron) based on MATLAB and collected environmental/vehicle data and triggered signal were dealt with input/output for training MLP, respectively. Then the output of MLP is quantitatively considered as driving suitability index to visualize how driving situation is appropriate or not. With this designed index, it can be used for designing threshold of controllers in ADAS (Advanced Driver Assistance System)/AV (Automated Vehicle) systems to consider various drivers’ acceptances.