Design of DDPG-Based Extended Look-Ahead for Longitudinal and Lateral Control of Vehicle Platoon
Anggera Bayuwindra, Leon Wonohito, Bambang Riyanto Trilaksono
Abstract
This paper presents a novel Deep Deterministic Policy Gradient (DDPG) algorithm with extended look-ahead approach for longitudinal and lateral control of vehicle platooning. The DDPG algorithm is adapted due to its ability to fit nonlinear system and to handle continuous control environment. Moreover, the dynamic input inversion is introduced to reduce domain of the action space from DDPG output. The existing look-ahead approach is considered as a cost-effective approach since it uses the available information from on-board sensors and is effective against the loss of lane markings. However, the approach is known to suffer from cutting-corner phenomenon. To address cutting-corners, we introduce the extended look-ahead approach and derive the true-local error states using the already available information from lidar and V2V communication. The robustness and performance of DDPG-based extended look-ahead controller is investigated by means of simulations and validated through experiments on a Donkey Car platform. The simulations and experiments with Donkey Car show that the DDPG-based extended look-ahead algorithm can provide an efficient control strategy for longitudinal and lateral maneuvers without the requirement of path information.