Autonomous Vehicle Logged-Data Simulation for Realistic ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training autonomous vehicle machine learning models face challenges in acquiring sufficient quantity and quality of training data to accurately represent a wide variety of driving conditions, as real-world data collection is insufficient and simulation data from video games lacks realism.
Innovation Solution
Generate simulation data from logged data of autonomous vehicles, incorporating actor types and motion behavior characteristics to create realistic simulation scenarios for training models, using a system comprising processors and memory to process and manipulate sensor data to enhance training data variety and quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If simulation data from video games is used for training, then the quantity of training data is increased, but the quality and realism of the data deteriorates
Solution Approach 1:
The patent creates a digital twin or copy of the real autonomous vehicle by mapping logged sensor data from the real vehicle to a virtual replica in the simulation environment. This copying approach preserves the high quality and realism of real-world data while enabling unlimited quantity of training scenarios through virtual replication and augmentation.
2Measurement precision
If more real-world data is collected to improve training quality, then the quality of training data is improved, but the time and resources required increase
Solution Approach 1:
The patent performs preliminary action by collecting and storing logged sensor data from the real autonomous vehicle during normal operation. This pre-collected data is then mapped to create simulation scenarios, eliminating the need for time-consuming real-world data collection for each training scenario while maintaining high data quality.
Solution Approach 2:
By creating virtual copies of real-world driving scenarios through data mapping to digital twins, the system generates unlimited training scenarios from a single set of real-world recordings, dramatically reducing the time required to produce diverse training data while preserving realism.
3Quantity of substance
If simulation environments are created to generate unlimited training data, then the quantity of training data is increased, but the accuracy of representing real-world conditions deteriorates
Solution Approach 1:
The patent maps logged sensor data from the real autonomous vehicle to create an accurate digital twin in the simulation environment. This copying process preserves the fidelity of real-world sensor measurements while enabling unlimited scenario generation, ensuring that simulation data accurately represents real-world conditions.
Solution Approach 2:
The system uses parameter changes by varying scenario conditions (weather, lighting, traffic patterns) in the simulation while maintaining the core sensor data mappings from real-world logs. This allows generation of diverse training scenarios that all remain grounded in actual real-world measurements, preserving reliability while increasing quantity.
Data Source
AI summary
Logged data from an autonomous vehicle is processed to generate augmented data. The augmented data describes an actor in an environment of the autonomous vehicle, the actor having an associated actor type and an actor motion behavior characteristic. The augmented data may be varied to create different sets of augmented data. The sets of augmented data can be used to create one or more simulation scenarios that in turn are used to produce machine learning models to control the operation of autonomous vehicles.