Autonomous Vehicle Lane Change Path Determination Using Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is no existing technology that effectively determines a lane change path for autonomous vehicles using deep learning, which is crucial for improving driving stability in dynamic and changing driving situations.
Innovation Solution
An apparatus and method utilizing deep learning to extract and determine a final lane change path based on properties such as collision risk, path curvature, and length, allowing the autonomous vehicle to optimize its lane change strategy in real-time by generating multiple lane change paths considering dynamic and static obstacles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple lane change paths are generated and evaluated using deep learning, then driving stability and safety are improved, but computational complexity and processing time increase
Solution Approach 1:
The lane change path determination is divided into multiple independent evaluation dimensions (collision risk, path curvature, path length) that are processed separately and then combined. This segmentation allows the system to evaluate multiple paths without overwhelming computational complexity, as each path can be assessed through standardized modular criteria.
Solution Approach 2:
The system pre-generates multiple candidate lane change paths before final selection, allowing the deep learning model to evaluate several options in advance. This preliminary action enables the system to choose the optimal path based on comprehensive evaluation of multiple factors, improving driving stability while managing computational load through efficient pre-processing.
2Manufacturing precision
If deep learning models process multiple lane change paths in real-time, then path optimization improves, but processing speed decreases
Solution Approach 1:
The system transforms the complex path optimization problem into evaluations of specific measurable parameters (collision risk, path curvature, path length). By changing the problem representation from abstract path optimization to concrete parameter comparison, the deep learning model can process multiple paths more efficiently while maintaining high optimization precision.
Solution Approach 2:
Different evaluation criteria are applied to different aspects of path quality locally - collision risk assessment focuses on safety-critical regions, path curvature evaluation addresses comfort requirements, and path length measurement optimizes efficiency. This localized quality assessment allows parallel processing of multiple paths with optimized computational resources.
3Measurement precision
If the system extracts at least two lane change paths based on deep learning, then decision accuracy improves, but system complexity increases
Solution Approach 1:
The system implements feedback mechanisms where the evaluated lane change paths and their associated metrics (collision risk, curvature, length) are fed back into the decision-making process. This feedback loop allows the system to refine path selection based on comprehensive evaluation results, improving decision accuracy while managing complexity through iterative optimization rather than overly complex architectures.
Data Source
AI summary
An apparatus for determining a lane change path of an autonomous vehicle is provided. The apparatus includes a learning device configured to learn lane change paths corresponding to a lane change strategy of the autonomous vehicle, and a controller configured to interwork with the learning device to extract at least two lane change paths corresponding to the lane change strategy among a plurality of lane change paths in a drivable area of the autonomous vehicle and determine a final lane change path based on properties of the extracted lane change paths.