New LSTM Framework Predicts Vehicle Intent at Intersections with High Accuracy

Autonomous vehicles require highly accurate intention prediction, especially in complex scenarios like intersections, to ensure safety and agility. A new framework, dubbed INTENT, has been introduced to address this critical need. INTENT leverages an LSTM (Long Short-Term Memory) model to forecast a vehicle's intended action—whether it will go straight, turn left, or turn right—at intersections, with a prediction window of two seconds before the event occurs. The framework's effectiveness was rigorously tested using the InD dataset, which captures real-world driving behaviors. Through extensive experiments and ablation studies, INTENT demonstrated exceptional performance, achieving an accuracy of 99.71%. This high level of precision is crucial for autonomous systems to make timely and correct evasive actions, minimizing potential damage in emergency situations and prioritizing safety. Beyond direct safety applications, intention prediction capabilities like those offered by INTENT can also significantly improve trajectory prediction, making autonomous vehicle navigation more robust and human-like in its interpretation of driving scenarios.