Hybrid BEV Sensor Fusion for Certifiable Vehicle Scene Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current advanced driver assistance systems (ADAS) face challenges in generating a comprehensive and versatile representation of the vehicle's surrounding environment, particularly in supporting all levels of driving automation and operational design domains, while ensuring safety and scalability.
Innovation Solution
A computing system that dynamically generates a sensor-fused, hybrid grid-based bird's eye view representation using both traditional geometric formulas and learned sensor data processing, combining raw data from various sensors like LIDAR, radar, and cameras to create a fused environment representation suitable for advanced driver assistance and autonomous driving tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If only traditional geometric reprojection methods are used to generate BEV grid maps, then the system ensures interpretability and safety certification, but the system lacks the ability to capture high-level environmental features and complex patterns
Solution Approach 1:
The patent merges traditional geometric reprojection methods with learned sensor data processing methods to generate a hybrid BEV representation. The traditional module ensures interpretability and safety certification through deterministic geometric transformations, while the learned module captures high-level environmental features and complex patterns through neural network processing of sensor data, thereby resolving the contradiction between reliability and adaptability.
2Adaptability or versatility
If only learned sensor data processing is used to generate BEV maps, then the system captures high-level features and complex patterns, but the system loses interpretability and becomes difficult to certify for safety
Solution Approach 1:
The patent combines learned sensor data processing with traditional geometric reprojection methods. The learned processing module captures high-level features and complex patterns while the traditional geometric module maintains interpretability and enables safety certification through deterministic, verifiable transformations. This hybrid approach resolves the contradiction between adaptability and reliability.
3Device complexity
If a single sensor data type is processed, then the system simplifies processing complexity, but the system fails to generate a comprehensive fused environment representation
Solution Approach 1:
The patent implements a universal processing framework that handles multiple sensor data types (LIDAR, radar, cameras) through a unified hybrid approach. The system processes each sensor type through both traditional geometric reprojection and learned processing methods, then fuses the results to generate a comprehensive environment representation that leverages the strengths of each sensor modality while maintaining manageable processing complexity.
4Adaptability or versatility
If multiple sensor data types are fused, then the system generates a comprehensive environment representation, but the system increases processing complexity and computational load
Solution Approach 1:
The patent merges multiple sensor data types through a hybrid processing architecture that applies both traditional geometric reprojection and learned processing methods in parallel. This approach enables comprehensive environment representation by fusing LIDAR, radar, and camera data while managing computational complexity through efficient integration of deterministic and data-driven methods.
Data Source
AI summary
A vehicle computing system can receive raw sensor data in both a traditional sensor data processing module and a learned sensor data processing module. Each module can reproject sensor data in BEV space, and can optionally perform sensor fusion when multiple sensor data types are processed. The system can then combine the learned BEV grid map or volume and the traditional BEV grid map or volume to generate a hybrid BEV representation of a surrounding environment of the vehicle, and process the hybrid BEV representation of the surrounding environment to derive a fused representation of the surrounding environment of the vehicle