Aborted lane-change strategy based on Gauss Mixture Hidden Markov model

The concept of aborted lane-change is introduced in this paper, and the lane-change decisions are divided into four states: lane-keeping, lane-change, keeping lane-change and aborting lane-change. Considering the lane change characteristics of actual drivers, an aborted lane-change model based on the Gauss mixture hidden Markov model (GMM-HMM) is proposed. Meanwhile, the concept of aggressiveness is proposed and calculated by Long Short-Term Memory (LSTM). The aggressiveness quantifies surrounding vehicle driving characteristics and is included as input to the lane-change decision model. Compared with the previous lane-change model, the simulation results show that the proposed model improves the correct rate to 95% after incorporating the scenario of aborted lane-change; with the consideration of aggressiveness, the correct rate is further improved to 97.5%, and the performance in online validation is well with errors less than 2s. It can be concluded that the model proposed in this paper can better simulate the driver's lane-change behavior, which is instructive for future work.