Autonomous Vehicle Camera Control With Safety-Value Gating
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomously moving vehicles or aircraft face challenges in navigating unpredictable environments, such as public spaces, due to high data volume requirements for detailed maps and difficulty in responding to dynamic obstacles and fellow road users, leading to potential navigation failures or unsafe performance when encountering untrained situations.
Innovation Solution
A system comprising a navigation module, camera, recognition module, safety-determining module, and control module that processes live camera images to determine navigation instructions and safety values, allowing the vehicle or aircraft to adjust control values and acceleration based on real-time conditions, using deep learning to recognize patterns and adapt driving behavior, and learn from operator inputs to improve safety and navigation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If detailed high-resolution maps and sensors are used to detect obstacles in unpredictable environments, then navigation safety is improved, but data volume requirements become unacceptably high
Solution Approach 1:
The system performs preliminary actions by training the autonomous vehicle with recorded camera images and operator control inputs before actual operation. During training, the system learns to associate visual patterns with appropriate navigation decisions, so that during actual operation, it can make safety decisions without requiring extensive real-time data processing or high-resolution maps
Solution Approach 2:
The system uses recorded camera images from training sessions as copies of real-world scenarios. These images are stored and reused during operation to match against current visual input, allowing the system to recognize patterns and make decisions based on previously encountered situations without requiring new high-resolution mapping data
2Adaptability or versatility
If the system is trained with operator control inputs to improve autonomous navigation, then adaptability to changing conditions is improved, but the system may still fail when encountering untrained situations
Solution Approach 1:
The system implements feedback by continuously comparing current camera images with stored training images and adjusting its navigation decisions based on the degree of correspondence. When the system encounters situations that do not match training data sufficiently, it can request additional operator input, which then becomes part of the training set for future improvements
Solution Approach 2:
The system dynamically adapts its operation mode based on the confidence level of its decisions. When encountering untrained situations, it transitions from fully autonomous operation to requesting operator input, and the training data is dynamically updated to include these new scenarios, making the system progressively more capable of handling previously untrained situations
3Reliability
If the system requests operator control for every untrained situation, then safety is improved, but navigation efficiency and productivity decrease
Solution Approach 1:
The system applies partial operator control only when necessary - specifically when the degree of correspondence between current images and training images falls below a threshold. For the majority of routine situations that match training data, the system operates autonomously without operator intervention, maintaining both safety and efficiency
Data Source
AI summary
System for controlling an autonomous vehicle on the basis of control values and acceleration values, having a safety-determining module configured to receive live images from a camera, to receive recorded stored images preprocessed for image recognition from an internal safety-determining module data storage, to receive navigation instruction(s) from a navigation module; to compare the live images with the stored images to determine a degree of correspondence; and to determine a safety value which indicates the extent to which the determined degree of correspondence suffices to execute the navigation instruction(s); wherein a control module is configured to receive the navigation instruction(s); to receive the live images; to receive the safety value; to compare the safety value to a predetermined value; and if the safety value is greater than the predetermined value, to convert the navigation instruction(s) and the camera images into control values and acceleration values for the autonomous vehicle.