Trajectory Jerking Suppression for Mixed Traffic Flow at a Signalized Intersection: A Trajectory Prediction Based Deep Reinforcement Learning Method

Abstract

Vehicles stopping at signalized intersections during a red light is one of the main causes of traffic oscillations. Recently, deep reinforcement learning (DRL) methods have been applied to connected and automated vehicles (CAVs) traffic to reduce the traffic oscillation at signalized intersections. However, these methods do not perform well for mixed traffic flow, including human vehicles (HVs) and CAVs, especially when the CAV rate is low. We found that this was because they did not take into account the HVs stopping at a red light and causing oscillations. If this oscillation is ignored during the speed regulation, CAVs may conflict with the oscillation wave in the future, forcing a sudden and significant speed reduction and triggering the so-called “trajectory jerking” phenomenon. In order to address this problem, this study proposes a trajectory prediction-based DRL method. By introducing the prediction of the downstream vehicle trajectory into the design of the reward function, the leading CAV will adjust its speed in advance to avoid the future oscillation wave caused by HV’s stopping during the red phase. Simulation tests on various penetration rates of CAVs are conducted for the mixed traffic environment to evaluate the performance of the proposed method. The results show that the proposed method has two advantages. Even with low CAV penetration, the oscillations are suppressed remarkably well. And fuel consumption is also significantly reduced. This research provides a new idea to suppress traffic oscillations in a mixed traffic environment.