Autonomous Trajectory Prediction Using Uncertainty-Optimized Waypoints

The patent segments the trajectory prediction process into multiple discrete waypoints along the predicted path. Each waypoint is associated with an uncertainty distribution that is optimized independently. This segmentation allows the system to handle long-term predictions by breaking them into manageable segments, where each segment's uncertainty is modeled and optimized separately, thereby maintaining accuracy over extended time horizons.

The patent transforms the trajectory representation from fixed points to probability distributions parameterized by mean and covariance. By optimizing the parameters of these uncertainty distributions (mean position, covariance matrix) at each waypoint, the system adapts to dynamic environmental changes and maintains prediction accuracy over longer time horizons, resolving the contradiction between prediction horizon and accuracy.

The patent applies optimization techniques selectively rather than uniformly across the entire trajectory. Optimization is focused on determining solution waypoints at critical segments of the trajectory where uncertainty is highest or where environmental dynamics are most significant. This partial application of optimization maintains accuracy improvements while limiting computational complexity increases.

The system uses the predicted trajectory and uncertainty distributions themselves to guide the optimization process. The uncertainty distributions at each waypoint provide self-contained information about where optimization efforts should be concentrated, allowing the system to automatically allocate computational resources to the most critical prediction segments without external intervention or complex centralized control.