Early Lane Change Prediction for Automated Driving Systems Using Multi-Task Attention-Based Convolutional Neural Networks

Lane change (LC) is one of the safety-critical manoeuvres in highway driving according to various road accident records. Thus, reliably predicting such manoeuvre advance critical for safe and comfortable operation automated systems. The majority previous studies rely on detecting a that has been already started, rather than advance. Furthermore, most works do not estimate key timings (e.g., crossing time), which can actually yield more useful information decision making ego vehicle. To address these shortcomings, this paper proposes novel multi-task model simultaneously likelihood LC time-to-lane-change (TTLC). In both tasks, an attention-based convolutional neural network (CNN) used as shared feature extractor from bird’s eye view representation environment. spatial attention CNN improves extraction process by focusing relevant areas surrounding addition, two curriculum learning schemes are employed train proposed approach. extensive evaluation comparative analysis method existing benchmark datasets show outperforms state-of-the-art prediction models, particularly considering long-term performance.