A Personalized Behavior Learning System for Human-Like Longitudinal Speed Control of Autonomous Vehicles

• DOI: 10.3390/s19173672
• Publication Date: 2019-08-26T08:38:23Z
• Authors: Chao Lu; Jianwei Gong; Chen Lv; Xin Chen; Dongpu Cao; Yimin Chen
• Keywords: Planner; Smoothness; Driving simulator

As the main component of an autonomous driving system, motion planner plays essential role for safe and efficient driving. However, traditional planners cannot make full use on-board sensing information lack ability to efficiently adapt different scenes behaviors drivers. To overcome this limitation, a personalized behavior learning system (PBLS) is proposed in paper improve performance planner. This based on neural reinforcement (NRL) technique, which can learn from human drivers online realize human-like longitudinal speed control (LSC) through demonstration (LFD) paradigm. Under LFD framework, desired be learned by PBLS converted low-level commands proportion integration differentiation (PID) controller. Experiments using simulator real data show that reproducing their LSC scenes. Moreover, comparative experiment with adaptive cruise (ACC) demonstrates superior maintaining comfort smoothness.