Agent Behavior Prediction for Vehicle Motion Planning

The patent segments the prediction task into two distinct components: trajectory prediction and behavior prediction. Instead of using a single complex model to predict detailed trajectories, the system uses separate models - one for predicting coarse trajectories and another for predicting behavioral intentions. This segmentation reduces the complexity of each individual model while maintaining overall prediction accuracy.

The patent extracts behavior prediction as a separate function from trajectory prediction. By taking out the behavior component and modeling it independently based on interaction patterns rather than detailed trajectory data, the system reduces the complexity of the trajectory prediction model while preserving the essential predictive information needed for safe navigation.

The patent divides the computational task into segmented prediction stages: first predicting trajectories, then predicting behaviors based on those trajectories and interaction patterns. This segmentation allows each stage to use computationally efficient models appropriate to its specific task, improving overall productivity while maintaining reliability through the combined output of both stages.

The patent implements dynamic behavior prediction that adapts to different interaction scenarios. Rather than using a static complex model for all situations, the system dynamically adjusts behavior predictions based on the specific trajectory patterns and interaction contexts observed, improving computational efficiency while maintaining reliability across diverse driving scenarios.

The patent introduces behavior prediction as an intermediary layer between trajectory prediction and motion planning. Instead of directly using trajectory predictions for motion planning, the system first predicts the behavior intentions of surrounding agents, then uses both trajectory and behavior information to inform motion planning decisions. This intermediary behavior layer provides richer contextual understanding, enabling less conservative motion planning while maintaining safety.

The patent performs preliminary behavior prediction before final motion planning decisions are made. By anticipating the behavioral intentions of surrounding agents in advance, the system can plan more aggressive and efficient maneuvers knowing the likely responses of other agents, rather than defaulting to conservative behavior to account for uncertainty.