Autonomous Driving Behavior Determination via Collision-Free Set Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving solutions based on neural networks prioritize human-like driving behaviors over safety, leading to potential risks during autonomous driving processes.
Innovation Solution
A method and device that utilize a trained neural network model and anomaly detection model to derive a driving behavior set with no collision risk, process driving data to obtain a predicted behavior, and ensure data normality before processing, enhancing safety by selecting reliable behaviors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a neural network model is used to learn human driving behavior features, then the driving behavior becomes more human-like, but safety is ignored and potential risks arise
Solution Approach 1:
The patent segments the driving behavior determination into two independent modules: a neural network model for generating human-like driving behaviors, and a derivation module for calculating collision-free driving behaviors. This segmentation allows each module to specialize in one function while the system as a whole achieves both human-likeness and safety through integration of both modules
Solution Approach 2:
The patent introduces an intermediary mechanism that compares the predicted driving behavior from the neural network with the driving behavior set derived through collision-free derivation. This intermediary comparison step acts as a safety filter, ensuring that only collision-free behaviors are executed while preserving the human-like characteristics of neural network predictions
2Device complexity
If the neural network model predicts driving behavior directly, then the system is simple and fast, but it cannot guarantee collision-free operations
Solution Approach 1:
The patent applies preliminary action by pre-calculating a driving behavior set through derivation that guarantees collision-free operations before the neural network makes its prediction. This pre-established safe behavior set serves as a constraint framework within which the neural network operates, ensuring safety is built-in rather than added as an afterthought
Solution Approach 2:
The patent introduces dynamics by allowing the system to adaptively select between different driving behaviors based on real-time conditions. The neural network dynamically predicts the most human-like behavior while the derivation module dynamically generates collision-free alternatives, and the system dynamically integrates these to achieve both safety and naturalness
Data Source
Figure 1A
Figure 1B~2
Figure 3~4
AI summary
This application provides a driving behavior determining method and a related device thereof, to avoid a potential risk of autonomous driving and improve a safety level of autonomous driving. The method in this application includes: obtaining first driving data of a first vehicle and second driving data of a second vehicle; obtaining, based on the first driving data and the second driving data, a driving behavior set in which the first vehicle has no risk of collision with the second vehicle; processing the first driving data and the second driving data by using a target model, to obtain a predicted driving behavior of the first vehicle; and if the driving behavior set includes the predicted driving behavior, determining the predicted driving behavior as a driving behavior to be performed by the first vehicle.