Autonomous Vehicles: Trajectory Prediction and Social Behavior

This lecture covers the challenges and solutions in deep learning for autonomous vehicles, focusing on vehicle behavior prediction and trajectory forecasting. It discusses the complexity of inputs, the importance of social behavior modeling, and the need for feasible and interpretable outputs. Various approaches and models, such as radar systems, map perception, and knowledge-aware models, are explored. The lecture emphasizes the significance of understanding human trajectory forecasting and the role of social anchors in predicting future trajectories. It also delves into scene-specific residuals, directional LSTMs, and the interpretability of intents in real-world scenarios.