Abstract
Vehicle intention prediction is a pivotal aspect in the agility and safety of autonomous vehicles in all driving scenarios, especially in cases requiring significant human interaction and complex driving behaviors like at intersections, roundabouts, and emergency situations such as sudden stops. Accurate intention prediction aids in taking the correct evasive action within a critical time frame where every second counts to prevent catastrophic incidents. In the worst-case scenario, it helps minimize damage and prioritize safety.
Intention prediction can also enhance trajectory prediction (intention conditioned trajectory prediction). This study proposes the INTENT framework using an LSTM model to predict the vehicle's intention at intersections 2 seconds ahead of the event occurrence—whether vehicles are going straight, turning left, or turning right. Comprehensive model experiments and ablation studies were conducted on the InD dataset, achieving an impressive accuracy of 99.71%.
Blogger's Review: The INTENT framework showcases the immense potential of deep learning in the realm of autonomous driving, particularly in enhancing safety and responsiveness in complex driving scenarios. As datasets and algorithms continue to improve, the accuracy and real-time capabilities of these predictions are expected to advance further.