Synthetic Sensor Data Generation for Autonomous Vehicle Scenario Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The development of control algorithms for autonomous vehicles is hindered by the limited nature of sample sensor data, which is often specific to particular environments and cannot effectively simulate unsafe conditions, making it difficult to ensure robust evaluation and testing of algorithm performance.
Innovation Solution
A system and method for generating synthetic sensor data, such as radar, lidar, and sonar data, from 3D scene data, allowing for the creation, evaluation, and verification of controller designs in a custom-designed environment, using a 3D scene generator, synthetic sensor data generator, and modeling environment that includes virtual sensors and ground truth data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real sensor data from specific environments is used for testing, then the data reflects actual operating conditions, but the data cannot simulate unsafe conditions and is limited to particular environments
Solution Approach 1:
The patent creates virtual copies of sensor environments through 3D scene generation. Virtual sensor data is synthesized from 3D scene data, allowing reproduction of various environments including unsafe conditions without physical presence. This copying approach enables comprehensive testing while maintaining reliability through controlled virtual scenarios.
Solution Approach 2:
The system performs preliminary generation of 3D scene data and virtual sensor data before actual algorithm testing. By pre-configuring diverse environments, including rare and unsafe conditions, the system prepares comprehensive test datasets in advance, enabling robust algorithm evaluation without requiring physical exposure to all scenarios.
2Adaptability or versatility
If virtual sensor data is generated from 3D scene data, then diverse and customizable scenarios can be simulated, but the complexity of the generation system increases
Solution Approach 1:
The 3D scene generator and virtual sensor data generator are designed as multi-functional systems that can produce various types of sensor data (radar, lidar, camera) from unified 3D scene representations. This universal approach consolidates multiple generation functions into integrated components, managing system complexity while maintaining high adaptability for diverse scenario customization.
3Measurement precision
If comprehensive 3D scene data is used to generate virtual sensor data, then realistic sensor simulations are achieved, but the computational resources and processing time increase
Solution Approach 1:
The system implements selective 3D scene generation by creating only the portions of scenes necessary for specific testing objectives. Rather than generating complete comprehensive scenes for all tests, the approach uses partial scene generation tailored to particular algorithm evaluation needs, reducing computational overhead while maintaining sufficient realism for targeted testing scenarios.
Data Source
AI summary
Systems and methods generate synthetic sensor data, such as synthetic radar, lidar, and/or sonar data from three dimensional (3D) scene data that may be custom designed. Reflectivity coefficients in the radar, lidar, and/or sonar spectrums may be determined for objects included in the 3D scene data. The reflectivity coefficients may be utilized by a game engine for computing the synthetic sensor data. The synthetic sensor data may be used in the creation, evaluation, and/or verification of a design for a controller or other system that utilizes such data.