Occlusion-Based Simulation Divergence for AV Object Visibility

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Autonomous vehicles face challenges in capturing sufficient real-world data for training machine learning models, particularly for rare scenarios like pedestrians unexpectedly crossing the street, due to their infrequent occurrence, which slows down the training process and affects model performance.

Innovation Solution

A simulation platform generates synthetic AV scene data by accurately simulating real-world scenarios, including occlusion regions to determine object visibility, which is validated against real-world data to ensure accuracy and improve training efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If real-world data is captured for training machine learning models, then model performance improves, but training speed decreases due to the infrequent occurrence of rare scenarios

Engineering Contradiction:

Improvemodel performanceVSAvoidtraining speed

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent creates synthetic copies of real-world driving scenarios through simulation. A simulation platform generates virtual training data that replicates rare events (like pedestrians unexpectedly crossing) without requiring actual real-world capture. This copying approach maintains model performance while dramatically accelerating training speed by unlimitedly reproducing rare scenarios in silico

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system performs preliminary validation of simulation accuracy by comparing synthetic occlusion regions against real-world data before using the simulation for training. This preliminary action ensures the simulation faithfully reproduces real-world conditions, allowing subsequent training to proceed at simulated speed while maintaining the reliability of real-world data

Inventive Principle:

Principle #10Preliminary action

2Productivity

If simulation is used to generate training data, then training speed improves, but data accuracy may deteriorate

Engineering Contradiction:

Improvetraining speedVSAvoiddata accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system implements a feedback mechanism where simulation-generated occlusion regions are validated against real-world sensor data. The comparison process provides feedback on simulation accuracy, allowing the system to verify that synthetic data maintains the measurement precision needed for reliable model training while enjoying the speed benefits of simulation

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12187312B2Measuring environmental divergence in a simulation using object occlusion estimation

Publication Date: 2025.01.07 GM CRUISE HOLDINGS LLC

AI summary

Aspects of the disclosed technology provide solutions for measuring divergence between recorded real-world scene data, and a simulated environment. A process of the disclosed technology can include steps for generating, based on real-world autonomous vehicle (AV) scene data, a computer-generated simulation of a real-world scenario, determining, an occlusion region in which at least one surrounding object obstructs a field of view of the AV, and determining, based on the occlusion region, a portion of the object of interest that is visible to the AV. In some aspects, the process can further include steps for determining a divergence value indicating a difference between the portion of the object of interest that is visible to the AV within the simulation of the real-world scenario to an actual portion of the object of interest that is visible to the AV within the real-world scenario. Systems and machine-readable media are also provided.