Unsupervised Learning for Automated Driving Simulation Scenarios
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current simulation methods for automated driving are inadequate as they only provide limited and time-consuming manual generation of scenarios, failing to cover the vast range of road conditions and test variations required for efficient evaluation of vehicle software for highly automated driving systems.
Innovation Solution
A method utilizing unsupervised learning on sensed vehicle data to generate simulation scenarios, including behavioral, classification, and environmental information, allowing for the creation of realistic and varied driving scenarios that adapt to changes in behavior and conditions, and can deviate from observed patterns to introduce new scenarios not covered by initial data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual generation of scenarios is used, then scenario accuracy and realism are improved, but time consumption and cost increase significantly
Solution Approach 1:
The patent creates virtual copies of real-world driving scenarios by capturing actual sensor data from vehicles and reconstructing these scenarios in a simulation environment. This allows realistic scenario generation without manual creation, as the system copies real driving conditions including sensor readings, environmental factors, and traffic patterns directly into virtual simulations.
Solution Approach 2:
The system enables self-service scenario generation by automatically processing raw sensor data from vehicles, extracting relevant driving scenarios, and generating simulation cases without human intervention. The automated pipeline includes data collection, scenario extraction, parameterization, and validation steps that occur autonomously, eliminating the need for manual scenario crafting while maintaining high realism.
2Reliability
If manual generation of scenarios is used, then scenario quality is improved, but productivity decreases due to limited scenario coverage
Solution Approach 1:
The patent implements a universal scenario generation system that processes diverse sensor data from multiple vehicle sources and generates various types of driving scenarios simultaneously. The system handles different road conditions, weather scenarios, traffic patterns, and edge cases through a single automated framework, enabling comprehensive scenario coverage across all driving conditions without requiring separate manual creation processes for each scenario type.
Solution Approach 2:
The system generates diverse scenarios by systematically varying parameters such as environmental conditions, traffic density, weather patterns, and road characteristics. By adjusting these parameters across wide ranges and combinations, the system creates numerous unique scenario variations from a single set of real-world sensor data, dramatically expanding scenario coverage while maintaining quality through parameter-based diversification.
3Adaptability or versatility
If extensive manual scenario generation is performed to cover all driving conditions, then scenario diversity is improved, but resource consumption increases
Solution Approach 1:
Instead of manually creating diverse scenarios from scratch, the system copies real-world sensor data that inherently contains natural diversity across different driving conditions. By reproducing these captured scenarios in virtual environments, the system achieves high scenario diversity without the resource-intensive manual creation process, as the diversity is already present in the collected sensor data from various real-world conditions.
Solution Approach 2:
The system performs preliminary data collection and scenario extraction in advance by capturing sensor data during normal vehicle operations. This pre-captured diverse scenario library is then reused and recombined for simulation testing, eliminating the need for repeated resource-intensive scenario generation. The preliminary action of data collection during regular driving provides a reusable foundation for diverse scenario creation.
Data Source
AI summary
A method for generating a simulation scenario, the method may include receiving sensed information that was sensed during driving sessions of vehicles; wherein the sensed information comprises visual information regarding multiple objects; determining, by applying an unsupervised learning process on the sensed information, driving scenario building blocks and occurrence information regarding an occurrence of the driving scenario building blocks; and generating the simulation scenario based on a selected set of driving scenario building blocks and on physical guidelines, wherein the generating comprises selecting, out of the driving scenario building blocks, the selected set of driving scenario building blocks; wherein the generating is responsive to at least a part of the occurrence information and to at least one simulation scenario limitation.