Inverse Model Transfer Learning for AV Virtual Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle (AV) testing in simulated environments is limited by the complexity of real-world conditions, leading to slow simulation times and inadequate exploration of the design space, which hampers the development of safe and efficient AVs due to varying regulatory requirements and the need for comprehensive testing.
Innovation Solution
An inverse model neural network is used to generate design parameters for AVs, augmented by transferring knowledge from pre-existing models, and visualized through 3D/2D scatter plots or parallel plots to facilitate expert evaluation, enabling efficient exploration of the design space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If comprehensive testing in simulated environment is performed to understand AV operating patterns and train AI model, then safety and reliability of AV is improved, but time consumption and computational resources increase significantly
Solution Approach 1:
The system performs preliminary actions by pre-generating diverse simulation scenarios and pre-training inverse models with transferred knowledge from existing designs. This allows the simulation to start with pre-prepared test cases and pre-established model structures, reducing the time needed during actual AV safety testing while maintaining comprehensive coverage of operating conditions
Solution Approach 2:
The system changes parameters by using inverse models to directly generate design parameters from desired safety outcomes rather than traditional forward simulation. This parameter inversion approach, combined with knowledge transfer from existing designs, enables faster exploration of the design space while achieving comprehensive safety testing coverage
2Measurement precision
If simulation complexity is increased to replicate real-world conditions accurately, then measurement precision and realism are improved, but simulation speed decreases
Solution Approach 1:
The system substitutes traditional forward physics-based simulation mechanics with inverse model-based parameter generation. Instead of running complex forward simulations to infer design parameters from outcomes, the inverse model directly maps desired safety metrics to optimal design parameters, maintaining accuracy while dramatically reducing computational time
Solution Approach 2:
The system creates simplified copies of complex real-world scenarios through pre-generated simulation cases and transferred knowledge models. These copied scenarios capture essential real-world conditions without requiring full physical simulation complexity, enabling faster yet still accurate design parameter generation
3Adaptability or versatility
If design space exploration is expanded to cover all possible AV configurations and scenarios, then adaptability and versatility are improved, but computational resources and time requirements increase
Solution Approach 1:
The system achieves universality through the inverse model framework that can handle multiple design configurations and scenario types through a single unified model structure. Knowledge transferred from existing designs enables the model to generalize across different AV configurations, providing broad design space coverage without requiring separate models for each scenario
Solution Approach 2:
The system efficiently explores expanded design space by changing approach from exhaustive forward simulation to inverse parameter generation. The inverse model directly computes optimal design parameters for diverse scenarios, and knowledge transfer from existing designs enables rapid adaptation to new configurations, maintaining high productivity even as design space coverage expands
Data Source
AI summary
A computer-implemented method of engineering design includes performing a simulation for a design under test of the design interacting in a real-world environment, in a computer processor, generating simulation data from performance of the simulation, storing the generated simulation data in a computer memory, extracting the generated simulation data to train a first inverse model neural network, and generating a plurality of design parameters from the inverse model neural network. A visualization representative of the generated plurality of design parameters for display to a user allows an expert user to evaluate the suggested design parameters and select some or all of the suggested design parameters. For existing designs more data is available for the simulation than for a new design. The inverse model for a new design may be augmented by transferring stored knowledge in a pre-existing inverse model to the model for the new design.