ANN Path Generation for Fast Autonomous Driving in Narrow Passages
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing path generation methods for autonomous driving, such as sampling-based algorithms, suffer from slow convergence speed, large memory requirements, and path generation delays, especially in narrow passages.
Innovation Solution
A processor-implemented method using an artificial neural network (ANN) model processes input data from recognition sensors and state data to generate path and control data simultaneously through a single forward process, utilizing quaternion or dual quaternion representations to efficiently output steering and acceleration data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If sampling-based algorithms are used for path planning, then path generation can be achieved, but convergence speed is slow and memory requirements are large
Solution Approach 1:
The patent replaces traditional sampling-based mechanical path planning algorithms with a neural network-based system. The neural network is trained offline to learn path planning patterns, and during runtime, it directly predicts control commands (steering angle, acceleration) based on current state and map data, eliminating the need for online sampling and iteration. This substitution dramatically reduces both convergence time and memory usage during operation.
Solution Approach 2:
The patent performs path planning learning in advance through offline training of the neural network. During the training phase, the system learns optimal path planning strategies by processing大量 training data consisting of map information and corresponding expert paths. This preliminary learning enables the network to make rapid predictions during actual operation without requiring complex real-time computations, thus solving the speed-memory contradiction.
2Productivity
If sampling-based algorithms are used for path planning, then path generation can be achieved, but path generation delay occurs in narrow passages
Solution Approach 1:
The patent replaces iterative sampling-based path planning with a direct neural network prediction system. The network takes current state and map data as input and directly outputs control commands without iteration, eliminating the time delays associated with sampling and convergence checks. This is particularly beneficial in narrow passages where sampling-based methods require many iterations to find valid paths.
Solution Approach 2:
The patent enables continuous path generation by using the neural network to predict control commands at every time step without interruption. The network processes current state and map data continuously, generating smooth and timely control outputs for steering and acceleration. This continuous operation eliminates the start-stop nature of sampling-based methods, improving productivity and reducing delays in dynamic environments.
3Ease of operation
If traditional path planning methods are used, then path generation can be achieved, but hyperparameter tuning is required
Solution Approach 1:
The patent implements self-service through automated hyperparameter optimization using Bayesian optimization. The system automatically searches for and identifies optimal hyperparameter configurations during the training phase without requiring manual intervention. This self-tuning capability simplifies operation by eliminating the need for users to manually adjust complex hyperparameters, while the neural network learns optimal parameters autonomously from training data.
Data Source
AI summary
A processor-implemented method with path generation includes obtaining input data that includes recognition sensor data and state data, inputting the input data into an artificial neural network (ANN) model and outputting output data corresponding to the input data in a single forward process, and obtaining path data and control data corresponding to the path data, based on the output data.