Self-Correcting Neural Network Controller for Autonomous Machines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks lack the ability to effectively recognize and correct incorrect output signals, and they do not have mechanisms to adaptively reduce mistakes with new information, leading to ineffective handling of errors and vulnerabilities to system malfunctions and virus attacks.
Innovation Solution
A controller system for autonomous machines that includes a first neural network trained with initial data and a detector to identify breaches of predetermined conditions, allowing for incremental re-training using local data, with features like data filtering and backup neural networks for secure operation and self-correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional neural networks are used for autonomous machine control, then the system can operate with simple architecture, but the system cannot recognize or correct incorrect output signals and is vulnerable to malfunctions
Solution Approach 1:
The controller is segmented into multiple functional modules: a neural network manager module that handles network operations, a detector module that monitors for triggering events, and a self-correction module that executes remediation actions. This modular segmentation allows each component to specialize in specific functions, improving overall reliability while keeping individual modules manageable in complexity.
Solution Approach 2:
The system performs preliminary actions by pre-configuring boundary conditions and triggering events before operation. The detector is pre-programmed with conditions that define incorrect output signals, and the neural network manager is pre-equipped with correction protocols. This preliminary preparation enables rapid response to errors without requiring complex real-time decision-making, thereby improving reliability without proportionally increasing operational complexity.
2Reliability
If neural networks operate without correction mechanisms, then the system maintains simple operation, but incorrect outputs cannot be identified or corrected
Solution Approach 1:
The system implements feedback through the detector module that continuously monitors neural network outputs against predefined boundary conditions. When incorrect outputs are detected (triggering events), the feedback loop activates the neural network manager to execute correction actions. This automated feedback mechanism improves output accuracy while maintaining operational simplicity by eliminating the need for manual monitoring and intervention.
Solution Approach 2:
The neural network manager autonomously detects triggering events and executes self-correction actions without external intervention. The system serves itself by automatically identifying incorrect outputs, retrieving appropriate correction data, and retraining the neural network. This self-service capability enhances output accuracy while preserving ease of operation, as the system manages its own errors without requiring complex external control procedures.
3Adaptability or versatility
If neural networks are trained only with initial data, then the training process is simple and fast, but the networks cannot adaptively reduce mistakes with new information
Solution Approach 1:
The training system transitions from static initial training data to dynamic adaptive training. The neural network manager continuously evaluates performance against boundary conditions and selectively incorporates new training data when triggering events occur. This dynamic approach allows the network to adapt to changing conditions and reduce mistakes over time, while the conditional nature of data incorporation prevents unnecessary complexity in the training process.
Solution Approach 2:
The system changes training parameters by selectively updating the training dataset based on detected triggering events. When incorrect outputs are identified, the system modifies the training parameters by incorporating relevant new data samples and adjusting training weights. This parameter-changing approach enables adaptive learning and improved network performance without requiring complete retraining or complex continuous optimization algorithms, thus balancing adaptability with manageable complexity.
4Object-affected harmful factors
If no boundary conditions are defined for neural network output, then the system operates with fewer constraints, but incorrect outputs cannot be prevented from affecting the autonomous machine
Solution Approach 1:
The system applies preliminary anti-action by pre-defining boundary conditions that represent incorrect or harmful outputs before the neural network operates. The detector is configured with these boundary conditions to proactively identify triggering events that indicate potential damage. By establishing these protective boundaries in advance, the system prevents harmful outputs from affecting the autonomous machine without requiring complex real-time analysis or intervention mechanisms.
Data Source
AI summary
Systems and methods for automatically self-correcting or correcting in real-time one or more neural networks after detecting a triggering event, or breaching boundary conditions are provided. Such a triggering event may indicate incorrect output signal or data being generated by the one or more neural networks. In particular, machine controllers of the invention limit the operations of neural networks to be within boundary conditions. Autonomous machines of the invention can be self-corrected after a breach of a boundary condition is detected. Autonomous land vehicles of the invention are capable of determining the timing of automatic transition to the manual control from automated driving mode. The controller of the invention filters and saves input-output data sets that fall within boundary conditions for later training of neural networks. The controllers of the invention include security architectures to prevent damages from virus attacks or system malfunctions.