Autonomous Vehicle U-Turn Path Generation Using Deep Learning Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle path generation systems, particularly in connected car services, limit the freedom of generating U-turn paths by relying on stored information and do not adequately consider real-time surrounding situations, leading to unnatural U-turn paths and a lack of adaptability in autonomous vehicles.
Innovation Solution
A deep learning-based method for generating U-turn paths in autonomous vehicles that calculates drivable areas, filters multiple paths using neural network learning, and determines an optimal path based on evaluation elements like risk, curvature, and length, incorporating real-time data from various sensors and infrastructure information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If existing vehicle path generation systems use stored information and expert knowledge from previous road tests, then the system complexity is reduced and ease of operation is improved, but the adaptability to real-time surrounding situations deteriorates and the path generation freedom is limited
Solution Approach 1:
The system pre-generates multiple candidate U-turn paths in advance before the vehicle actually needs to make a U-turn. These candidate paths are stored and can be quickly retrieved when needed, combining the efficiency of pre-computation with the flexibility of having multiple options to choose from based on real-time conditions
Solution Approach 2:
The system dynamically selects from multiple pre-generated candidate paths based on real-time surrounding situations. Rather than using a single fixed path, the system adapts its path selection by evaluating current traffic conditions, obstacles, and environmental factors to choose the most appropriate path from the candidate set
2Device complexity
If the system generates only one deterministic U-turn path based on stored expert information, then the device complexity is reduced, but the reliability of the path in complex real-time situations deteriorates
Solution Approach 1:
The path generation process is segmented into multiple independent candidate paths rather than producing a single path. Each candidate path represents a different routing option, and the system evaluates and selects the most appropriate one based on current conditions, thereby improving reliability without requiring a single overly complex path generation algorithm
3Reliability
If the system filters multiple paths using deep learning classification, then the adaptability and reliability of path selection is improved, but the device complexity and computational requirements increase
Solution Approach 1:
The system generates more candidate paths than will ultimately be needed (excessive action), then uses deep learning to filter down to the optimal path. This approach allows the system to explore a broader solution space and find better paths while managing computational complexity through selective filtering rather than optimizing a single path from scratch
Data Source
AI summary
A method for generating a U-turn path in an autonomous vehicle includes calculating a drivable area, generating multiple paths drivable in the drivable area, filtering a driving strategy path among the multiple paths based on deep learning, and determining a final path from the filtered candidate paths.