Autonomous Parking Path Planning With CNN Path Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving systems face challenges in efficiently planning and adapting real-time motion paths, particularly in environments with unknown map information and obstacles, due to slow convergence speed and high memory consumption of existing sampling-based algorithms.
Innovation Solution
A path distribution estimation model using a convolutional neural network (CNN) is trained to output path distributions based on initial information, including map and departure information, which allows for real-time path planning and adaptation by sampling probability distributions of optimal paths, incorporating unknown grid maps and occupancy grid maps, and utilizing sensors for continuous updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sampling-based algorithms are used for path planning in autonomous driving, then path finding capability is improved, but convergence speed deteriorates (slow convergence)
Solution Approach 1:
The system pre-calculates and stores multiple candidate paths along with their probability distributions before actual navigation. When the autonomous vehicle needs to navigate, it directly retrieves pre-computed paths rather than performing real-time sampling, significantly reducing convergence time while maintaining path finding reliability
Solution Approach 2:
The path distribution estimation model dynamically adapts to different environmental conditions by adjusting probability distributions based on map information, obstacle positions, and vehicle state. This allows the system to maintain high convergence speed while reliably finding appropriate paths for various scenarios
2Reliability
If sampling-based algorithms are used for path planning in autonomous driving, then path finding capability is improved, but memory consumption increases
Solution Approach 1:
The system extracts only the essential path information and probability distributions from complex sampling calculations, storing pre-computed results in a condensed format. This reduces memory consumption by eliminating redundant computational data while preserving the core path finding capability
Solution Approach 2:
The system transforms the representation of path data by using probability distribution parameters instead of storing complete sampling results. This parameter-based representation significantly reduces memory requirements while maintaining the ability to reliably find and evaluate paths
3Device complexity
If traditional path planning methods are used, then system complexity is reduced, but adaptability to dynamic environments deteriorates
Solution Approach 1:
The system replaces traditional mechanical path planning algorithms with a data-driven path distribution estimation model. This model uses learned probability distributions to adapt to dynamic environments automatically, providing high adaptability while maintaining manageable system complexity through the use of standardized deep learning frameworks
Data Source
AI summary
A device and method with continuous real-time autonomous parking planning and control are disclosed. A method of training a path distribution estimation model includes: obtaining initial information including map information, departure information of an autonomous driving device, and destination information of the autonomous driving device; obtaining paths through which the autonomous driving device is movable by inputting the initial information to the path distribution estimation model which predicts the paths; and training the path distribution estimation model to output a path distribution corresponding to the paths based on the initial information.