Vehicle Platooning Control with DDPG

This paper presents a controller design of vehicle platooning using an end-to-end Deep Deterministic Policy Gradient (DDPG) combined with Extended Look Ahead approach. The purpose of the combination is to maintain the expected result from the end-to-end DDPG system since Reinforcement Learning might be able to create unexpected result that considered optimal by the model but not by the user. In this paper, we use the extended look ahead approach as a guide for the DDPG system to learn avoiding a cutting comer behavior where the following vehicle turn too early and able to bypass the leader. To check the validity of the DDPG model, we use the car model from Donkey Car RC Simulator. The result shows that the following car is able to follow the leading vehicle without the cutting comer behavior and within acceptable distance.