Autonomous Vehicle Simulation Replay for Maneuver Validation

The patent creates virtual copies of real-world driving environments by reconstructing three-dimensional scenes from two-dimensional image data captured in actual driving scenarios. These synthesized virtual environments serve as testbeds for comprehensive software validation without requiring additional real-world data collection, thus resolving the contradiction between needing reliable validation and limited test data availability.

The system performs preliminary reconstruction of driving environments and pre-processing of test scenarios before actual software validation. By preparing virtual test environments in advance with diverse scenarios (including edge cases), the system enables comprehensive reliability testing without being constrained by the limited availability of real-world test data during the validation phase.

The system implements a hierarchical simulation approach where critical safety scenarios are simulated in full detail while less critical scenarios use simplified models. This partial action strategy achieves comprehensive testing coverage by focusing computational resources on the most important test cases, balancing testing completeness with acceptable computational time requirements.

The simulation process is divided into multiple independent stages: environment reconstruction, scenario generation, software execution, and result analysis. Each stage can be processed separately and parallelized, improving overall testing productivity while managing computational time through efficient resource allocation across different simulation segments.