AI Sensor Data Modification for Autonomous Driving Test Scenarios

The patent introduces an AI-based data modification system as an intermediary between simulation data and real-world testing requirements. The system takes simulation data as input and applies learned transformations (from real data pairs) to generate modified simulation data that bridges the gap between synthetic and real-world characteristics, thereby maintaining cost efficiency while improving realism.

The patent transforms simulation data by changing its parameters and characteristics through AI-based modification. By learning the parameter distributions and relationships from real data pairs and applying them to simulation data, the system alters the simulation data's parameters (such as sensor noise characteristics, object appearance, environmental conditions) to make them more representative of real-world scenarios.

The patent creates a virtual copy of real-world data characteristics by training an AI model on pairs of real simulation data and real-world data. The learned model captures the essential features and patterns of real data, which are then applied to generate modified simulation data that copies the realistic characteristics without requiring additional expensive real-world data collection campaigns.

The patent develops a universal AI-based modification system that can transform simulation data across multiple scenarios, vehicle types, and environmental conditions using a single trained model. This multi-functional approach allows the system to generate realistic test data for various ODDs (Operational Design Domains) without requiring separate real data campaigns for each specific scenario.

The patent implements a self-service system where the AI model automatically adapts simulation data to specific ODD details without requiring manual intervention. The system self-adjusts by applying the learned transformations from real data pairs, automatically handling scenario adaptation, parameter adjustment, and data generation based on the input specifications, thereby reducing complexity and time consumption.

The patent performs preliminary action by pre-training the AI model on comprehensive real data pairs that cover various ODD scenarios before actual data generation. This preliminary training phase enables the model to internalize the relationships between different ODD characteristics and their corresponding data patterns, so that subsequent adaptations to specific ODD details can be performed automatically without manual tuning.