Autonomous Vehicle Logged-Data Simulation for Realistic ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicle technology faces challenges in acquiring sufficient and high-quality training data to accurately represent various driving conditions, as real-world sensor data is insufficient and simulation data from video game-like simulators lacks realism.
Innovation Solution
Generating simulation data from logged sensor data of autonomous vehicles, which includes modifying actor characteristics and scenarios to create realistic simulation scenarios for training machine learning models used in perception, planning, and control subsystems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If simulation data from video game-like simulators is used to train machine learning models, 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 capturing sensor data from the actual vehicle and using it to generate simulation scenarios. This copying approach preserves the realism and accuracy of real-world data while enabling unlimited replication for training purposes, thus solving the contradiction between data quantity and data quality
Solution Approach 2:
The system varies simulation parameters such as weather conditions, lighting, actor behaviors, and sensor noise levels based on real-world logged data to generate diverse training scenarios. By changing parameters systematically while maintaining the underlying realism of the source data, the system produces both high-quality and high-quantity training data
2Reliability
If more real-world sensor data is collected to improve training data quality, then the accuracy of machine learning models is improved, but the time and resources required for data collection increase
Solution Approach 1:
The system performs preliminary action by capturing and storing sensor data from the real autonomous vehicle during normal operation. This pre-captured data is then used to generate extensive simulation scenarios without requiring additional real-world driving time, thus improving model quality while minimizing data collection time
Solution Approach 2:
The system enables continuous generation of training data from the pre-captured real-world sensor data through simulation. Instead of continuously collecting real-world data which is time-consuming, the system continuously generates diverse training scenarios from the existing data pool, maintaining high data quality while eliminating time loss
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.