From Observation to Prediction: LSTM for Vehicle Lane Change Forecasting on Highway On/Off-Ramps

TL;DR

The paper tackles predicting vehicle lane-change maneuvers at highway on/off-ramps (AoI) where variability is higher than on straight segments. It adopts a stacked LSTM framework to forecast lane-change events using ExiD and HighD datasets, comparing End-to-End and Multi-LSTM architectures and exploring different prediction horizons up to 4 seconds. Key contributions include an AoI-specific data processing pipeline, a systematic comparison of end-to-end vs modular learning, a horizon-informed evaluation, and a feature-bias analysis to identify the most influential inputs. The results indicate strong prediction performance on AoI up to 4 seconds and highlight trade-offs between accuracy and computational resources, with End-to-End offering favorable speed and efficiency, and future work proposing longer horizons and generative path modeling for extended scenario prediction.

Abstract

On and off-ramps are understudied road sections even though they introduce a higher level of variation in highway interactions. Predicting vehicles' behavior in these areas can decrease the impact of uncertainty and increase road safety. In this paper, the difference between this Area of Interest (AoI) and a straight highway section is studied. Multi-layered LSTM architecture to train the AoI model with ExiD drone dataset is utilized. In the process, different prediction horizons and different models' workflow are tested. The results show great promise on horizons up to 4 seconds with prediction accuracy starting from about 76% for the AoI and 94% for the general highway scenarios on the maximum horizon.

Figures (5)

• Figure 1: Area of Interest
• Figure 2: End-to-End Model
• Figure 3: Multi-L Models
• Figure 4: Survey Outcome
• Figure 5: Vehicle path example vs Prediction over time.