IMPLEMENTATION OF BEHAVIOURAL CLONING BASED ON DEEP NEURAL NETWORK FOR AUTONOMOUS CAR IN VIRTUAL ENVIRONTMENTS

• Main Author: Darfyma Putra, M.
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/47985
• Institution: Institut Teknologi Bandung

Behavioral cloning is a technique that applies the Deep Neural Network (DNN) method to mimic the behavior of drivers in driving a car. One of the behaviors that imitated is the decision to determine the steering angle of the car determined by the driver. Behavioral cloning replaces it with a prediction based on the driver's driving habits that have been studied previously so that the car can run autonomously. In this study an investigation will be conducted by observing the behavior of the model. The model is applied to the car and tested to obtain information on whether the technique is worthiness to use and safe for the car. This worthiness information will be used for future purposes as consideration or evaluation for further development. The test uses two types of experiments. In the first experiment the model will be observed based on the comparison of two driver sources to find out whether the Behavior Cloning technique can work well in imitating the behavior of the driver. A driver will act as an example of a main model that has safe driving characteristics, and then another person as a comparison will act the other way (ie driving unsafe). The second test uses the main character to ensure that the model can avoid other cars properly. Based on the results of two test scenarios, Behavioral cloning has been successfully implemented and is worth for use in autonomous cars. The first test results show the success of the model in imitating the character of each driver with MAE of 0.0177 and MSE of 6.6686x10-4. In the second test result shows the success of the model avoiding other drivers with MAE of 0.0228 and MSE of 1.0395x10-3.