Driving Behavior Simulation Using Scenario-Based Vehicle Interactions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computer models of driving behavior fail to reproduce realistic interactions between vehicles, leading to potential dangerous maneuvers and accidents in autonomous driving systems, especially in complex traffic scenarios like merging lanes or on/off-ramps, due to their reliance on non-empirical parameters.
Innovation Solution
A system utilizing environmental sensors and a machine learning-based driving behavior simulator that extracts observed vehicle behavior data from sensor data, adjusts driving behavior model parameters to minimize deviations between observed and simulated vehicle interactions, employing techniques like inverse reinforcement learning, reinforcement learning, and generative adversarial imitation learning to generate realistic driving behavior models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional computer models use non-empirical parameters, then the models are simpler to implement, but they fail to reproduce realistic driving behavior of interacting vehicles
Solution Approach 1:
The patent transforms driving behavior model parameters from non-empirical conventional values to empirical values derived from real sensor data of interacting vehicles. The processor extracts observed vehicle behavior data from sensor data and uses this empirical data to determine and adjust model parameters, fundamentally changing the parameter source from theoretical to observational, thereby achieving realistic driving behavior reproduction.
Solution Approach 2:
The patent creates a virtual copy of real driving interactions by generating simulated vehicle behavior data that mirrors observed vehicle behavior data from sensor recordings. The machine learning-based simulator reproduces the actual driving patterns, trajectories, and interactions of real vehicles, allowing autonomous systems to train on realistic scenarios without requiring physical test tracks.
2Reliability
If conventional computer models are used for training autonomous driving systems, then training can be performed with simpler data requirements, but the autonomous vehicle cannot interact properly with vehicles on public roads
Solution Approach 1:
The patent extracts specific useful information from large volumes of sensor data by using a processor to identify and extract observed vehicle behavior data from raw sensor recordings. The system selectively extracts relevant interaction patterns, trajectories, and behavioral characteristics from the comprehensive sensor data, converting massive raw data into concentrated training-relevant information.
Solution Approach 2:
The patent introduces a machine learning-based driving behavior simulator as an intermediary between real sensor data and autonomous vehicle training. The simulator acts as a mediator that processes observed vehicle behavior data and generates simulated training scenarios, bridging the gap between real-world data collection and virtual training environments, thereby enabling proper interaction capability development.
3Measurement precision
If sensor data from multiple environmental sensors is processed to extract observed vehicle behavior data, then the realism of the computer model improves, but the processing complexity and computational requirements increase
Solution Approach 1:
The patent merges data from multiple environmental sensors (radar, LIDAR, cameras) into unified observed vehicle behavior data. The processor integrates information from these different sensor types to create a comprehensive view of vehicle interactions, combining their respective strengths to achieve accurate and realistic behavior observation despite the increased processing complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An system 100 for determining a computer model of realistic driving behavior comprises one or more environmental sensors 110, which are configured to provide sensor data of observed vehicles 140-1 and 140-2 within a traffic area 150. A processor 120 is configured to extract observed vehicle behavior data from the sensor data and to extract driving behavior training data 310 from observed vehicle behavior data of vehicles interacting with each other within the traffic area 150. The system 100 is further using a machine learning based driving behavior simulator 130, which is configured to adjust driving behavior model parameters 320. The driving behavior model parameters 320 can be determined by reducing a deviation between the driving behavior training data 310 and simulated vehicle behavior data 340, which is generated by the machine learning based driving behavior simulator 130.