Autonomous Vehicle Path Planning With Offline-Learned Driving Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving technologies face challenges in efficiently generating driving paths within limited computing resources, leading to long calculation times that hinder real-time decision-making.
Innovation Solution
A method and apparatus for controlling autonomous driving that utilize machine learning to generate optimal driving paths based on training data, including information about drivable areas, and update parameters using a driving path loss calculation to improve path generation efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional path generation methods are used, then path generation accuracy can be maintained, but calculation time becomes excessively long
Solution Approach 1:
The system pre-generates a comprehensive database of driving paths covering various driving scenarios and environments before actual autonomous driving operations. This preliminary path generation database is stored and reused during real-time driving, eliminating the need for lengthy calculations during critical decision-making moments. The pre-computed paths are matched to current driving conditions through pattern recognition rather than recalculation.
Solution Approach 2:
Instead of generating new paths through complex real-time calculations, the system creates copies of pre-generated optimal paths from the database that match current driving conditions. The machine learning model identifies similar scenarios from historical data and replicates successful path solutions, significantly reducing computation time while maintaining path quality.
2Measurement precision
If machine learning models are trained with comprehensive data, then path generation accuracy improves, but computing resource requirements increase
Solution Approach 1:
The system extracts and separates the heavy computational training process from the real-time operating system. Comprehensive machine learning training is performed offline using substantial computing resources to build the path generation database and train models. During actual autonomous driving, only lightweight inference operations are executed, extracting predictions from the pre-trained models without requiring the full computational power used during training.
Solution Approach 2:
All intensive machine learning training and model optimization are completed in advance before deployment. The system performs preliminary actions of data processing, model training, and validation offline, storing the results in optimized formats that require minimal processing during real-time operations. This shifts the computational burden from the moving vehicle to pre-deployment infrastructure.
Data Source
AI summary
A method performed by a vehicle for controlling autonomous driving of the vehicle is introduced. The method may include steps of obtaining training data comprising information about a first drivable area, outputting, based on the information about the first drivable area, a first driving path in the first drivable area, determining, based on the first driving path and the training data, a driving path loss, updating, based on the driving path loss, at least one of parameters associated with the outputting the first driving path, obtaining, based on the updating the at least one of parameters, test data comprising information about a second drivable area for testing, outputting, based on the information about the second drivable area for testing, a second driving path for testing in the second drivable area for testing, and controlling, based on the second driving path for testing, autonomous driving of the vehicle.