車両ダイナミクスを考慮した深層強化学習による追従制御の再設計
Longitudinal Dynamic versus Kinematic Models for Car-Following Control Using Deep Reinforcement Learning
点質量モデルで学習した強化学習ACC制御器が車両ダイナミクス下で性能劣化することを示し、遅延入力と実加速度を状態に加える再設計で動的計画法に近い性能を達成した。
著者: Yuan Lin, John McPhee, Nasser L. Azad
分類: cs.RO, cs.LG, cs.SY, eess.SY
原文アブストラクト
The majority of current studies on autonomous vehicle control via deep reinforcement learning (DRL) utilize point-mass kinematic models, neglecting vehicle dynamics which includes acceleration delay and acceleration command dynamics. The acceleration delay, which results from sensing and actuation delays, results in delayed execution of the control inputs. The acceleration command dynamics dictates that the actual vehicle acceleration does not rise up to the desired command acceleration instantaneously due to dynamics. In this work, we investigate the feasibility of applying DRL controllers trained using vehicle kinematic models to more realistic driving control with vehicle dynamics. We consider a particular longitudinal car-following control, i.e., Adaptive Cruise Control (ACC), problem solved via DRL using a point-mass kinematic model. When such a controller is applied to car following with vehicle dynamics, we observe significantly degraded car-following performance. Therefore, we redesign the DRL framework to accommodate the acceleration delay and acceleration command dynamics by adding the delayed control inputs and the actual vehicle acceleration to the reinforcement learning environment state, respectively. The training results show that the redesigned DRL controller results in near-optimal control performance of car following with vehicle dynamics considered when compared with dynamic programming solutions.
