Closed-Loop Driving Scenario Simulation for AV Stack Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for testing autonomous driving systems (ADS) and advanced driver-assist systems (ADAS) rely heavily on physical world testing, which is expensive and time-consuming. Additionally, simulating all possible driving scenarios in the physical world is not viable, highlighting the need for advanced simulation techniques that can replicate real-world driving conditions effectively.
Innovation Solution
A computer system and method for testing autonomous vehicle (AV) stacks in simulation, which involves extracting driving scenarios from real-world data, inferring the goals or behaviors of real-world agents, and creating closed-loop simulations where both the ego agent and non-ego agents exhibit autonomous behavior. This system includes a trace extraction component, a driving scenario extraction component, a simulator, agent decision logic, and a test oracle to evaluate the performance of the AV stack.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical world testing is used to test autonomous driving systems, then testing accuracy and realism are improved, but testing cost and time consumption increase significantly
Solution Approach 1:
The patent creates virtual copies of real-world driving scenarios by extracting traces from physical world data and reconstructing them in a simulation environment. Non-ego agents are instantiated as virtual entities that replicate real-world behavior patterns, allowing comprehensive testing without physical deployment. This copying approach maintains testing realism while eliminating the time and resource costs of actual road testing.
2Adaptability or versatility
If all possible driving scenarios are simulated in the physical world, then scenario coverage is improved, but resource consumption and time requirements become unmanageable
Solution Approach 1:
The system performs preliminary extraction and analysis of real-world driving traces before simulation. By pre-processing physical world data to identify key scenario patterns, behavior models, and trace characteristics, the system prepares comprehensive test scenarios in advance. This preliminary action enables the simulation to cover diverse driving situations without requiring exhaustive physical testing of each scenario.
Solution Approach 2:
The simulation platform serves multiple testing functions simultaneously - it can test perception algorithms, decision-making systems, control strategies, and edge cases all within the same virtual environment. The extracted trace data and non-ego agent models are universally applicable across different testing objectives, allowing comprehensive scenario coverage without proportionally increasing resource requirements.
3Device complexity
If simple trace replay is used in simulation, then implementation complexity is reduced, but behavioral realism and closed-loop interaction are compromised
Solution Approach 1:
Instead of static trace replay, the system implements dynamic non-ego agents that can adapt their behavior during simulation. These agents use extracted behavior models to make real-time decisions based on simulated sensor inputs and environmental conditions. The agents' trajectories and actions are dynamically adjusted to maintain behavioral realism while responding to ego vehicle actions, creating authentic closed-loop interactions without excessive complexity.
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
A driving scenario is extracted from real-world driving data captured within a road layout. A simulation is run based on the extracted driving scenario, in which an ego agent and a simulated non-ego agent each exhibit closed-loop behaviour. The closed-loop behaviour of the ego agent is determined by autonomous decisions taken in an AV stack under testing in response to simulated inputs, reactive to the simulated agent. The closed-loop behaviour of the non-ego agent is determined by implementing an inferred goal or behaviour, reactive to the ego agent. The goal or behaviour is inferred from an observed trace of a real-world agent extracted from the real-world driving data.