Autonomous Vehicle Blind-Region Planning With Virtual Objects

The system creates virtual copies of potential objects (vehicles, pedestrians, cyclists) in blind spot regions based on probabilistic predictions. These virtual objects allow the planning system to simulate and evaluate trajectories as if the actual objects were present, enabling safer decision-making without requiring the vehicle to stop unnecessarily. The virtual objects are dynamically created and removed based on sensor updates and prediction accuracy.

The system performs preliminary probabilistic prediction and virtual object creation in advance before the vehicle reaches the blind spot region. By pre-calculating likely object positions and creating virtual representations, the system prepares multiple candidate trajectories ahead of time, allowing for smoother, more efficient navigation through uncertain regions rather than reacting with abrupt stops when uncertainty is detected.

The system dynamically adjusts the number, position, and properties of virtual objects based on changing environmental conditions, sensor data quality, and predicted risk levels. Rather than using fixed hard-coded rules for all blind spot scenarios, the system modulates virtual object parameters (such as probability weights, spatial distribution, and persistence) to match the actual uncertainty level, reducing computational overhead when uncertainty is low while maintaining safety when uncertainty is high.

The system transitions from static hard-coded blind spot responses to dynamic virtual object creation and manipulation. Virtual objects are continuously created, updated, and removed based on real-time sensor inputs and prediction accuracy, allowing the system to adapt computational resources to actual needs. This dynamic approach replaces rigid if-then rules with flexible probabilistic reasoning that scales computational effort according to situation complexity.