Simulation Recording Generation for Realistic ADS Collision Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated driving systems face challenges in being tested effectively due to the incompatibility of real collision videos with testing devices, which can distract from performance and pose risks of damage or injury from actual collisions.
Innovation Solution
A system is developed to produce simulation recordings by determining feature vector similarities between real collision videos and prospective recordings, using video language models to convert real videos into textual descriptions, and generating simulation recordings suitable for testing automated driving systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real collision videos are used to test automated driving systems, then the testing realism is improved, but the risk of damage or injury increases and the videos are incompatible with testing devices
Solution Approach 1:
The patent creates simulation recordings that replicate real collision scenarios without using actual collision footage. A perception system captures real collision data, which is then processed by a language model to generate textual descriptions. These descriptions are used to create synthetic simulation recordings that preserve the essential characteristics of real collisions while eliminating the harmful elements, allowing safe testing of automated driving systems.
2Measurement precision
If real collision videos are used for testing, then the training data quality is improved, but the incompatibility with testing devices prevents effective use
Solution Approach 1:
The patent introduces a language model as an intermediary between real collision videos and testing devices. The perception system extracts data from real videos, the language model converts this data into structured textual descriptions, and these descriptions are then used to generate simulation recordings that are compatible with automated driving system testing platforms. This intermediary process bridges the compatibility gap while preserving training data quality.
3Adaptability or versatility
If simulation recordings are created without using real video data, then compatibility with testing devices is improved, but the quality and realism of training data deteriorates
Solution Approach 1:
The patent performs preliminary processing of real collision video data before generating simulation recordings. The perception system first captures and analyzes real collision scenarios, extracting essential features and characteristics. This pre-processing ensures that the subsequent simulation generation is based on authentic collision data, maintaining high training data quality while achieving device compatibility through the simulation format.
Data Source
AI summary
A system for producing a simulation recording to test an automated driving system can include a processor, a communications device, and a memory. The memory can store a comparison module, a feedback module, and a communications module. The comparison module can determine a similarity between a feature vector associated with a first prospective video and a feature vector with a real video. The feedback module can cause, in response to the similarity being less than a threshold, feedback to be sent to a video language model to be used to convert the real video into a textual description to produce a second prospective recording. The communications module can cause, in response to the similarity being greater than the threshold, the first prospective recording to be communicated, via the communications device, to a device to test the automated driving system.