Neural Network Vehicle Control via Cost-Sensitive Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving systems face challenges in accurately classifying sensor data while considering safety aspects and adapting to varying environmental conditions, leading to potential misclassifications that can impact vehicle control, especially in situations where different errors have varying severity levels.
Innovation Solution
The method involves using an artificial neural network to classify sensor data based on both the network's output and environmental variables, with a cost matrix and decision rule that prioritizes safety by assigning higher costs to critical misclassifications, allowing for context-specific adjustments without retraining the model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cost-sensitive training is applied to penalize certain perceptual errors more severely, then safety-related aspects are taken into account, but convergence is difficult to guarantee and complete retraining is required for adjustments
Solution Approach 1:
The patent segments the classification process into two independent components: (1) the trained neural network model that provides base classification, and (2) a post-processing decision rule that applies cost-sensitive adjustment. This segmentation allows the model structure to remain fixed while enabling flexible, context-aware decision making through the separately configurable cost matrix and decision rule, avoiding the need for retraining when safety priorities change.
Solution Approach 2:
The patent introduces an intermediary decision rule that acts as a mediator between the neural network output and the final classification decision. This intermediary layer applies the cost matrix to adjust predictions based on safety considerations without requiring changes to the trained model itself, enabling flexible adaptation to different safety requirements through configuration rather than retraining.
2Reliability
If uncertainty assessment methods are used to indicate incorrect classification, then safety indication is provided, but computational intensity is very high for automotive control units
Solution Approach 1:
The patent employs a computationally inexpensive decision rule that operates on the neural network output without requiring intensive additional computation. Unlike methods such as Monte Carlo dropout that require multiple forward passes, this approach uses a simple cost matrix application that is computationally lightweight and suitable for real-time automotive control units with limited processing power.
3Reliability
If an ensemble of artificial neural networks is used to increase classification reliability, then correct classification is improved, but computational load and energy consumption become unrealistic for practical automotive applications
Solution Approach 1:
The patent merges the cost-sensitive decision making capability into a single neural network model through post-processing, rather than combining multiple separate network models. This consolidation achieves the reliability benefits of ensemble methods (through cost-aware decision adjustment) while avoiding the computational burden of running multiple networks in parallel, making it practical for automotive applications.
4Measurement precision
If different artificial neural networks are used for different contexts with switching between models, then contextual accuracy is improved, but computational effort increases significantly and training data requirements increase
Solution Approach 1:
The patent creates a universal decision rule framework that can handle multiple contexts and safety requirements through a single cost matrix configuration, rather than requiring separate trained models for each context. The same neural network model serves all contexts, with contextual adjustments applied through the configurable cost matrix and decision rule, reducing both computational overhead and training data requirements.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
To improve safety in autonomous or semi-autonomous driving, the invention provides a method for automatically executing a control function of an autonomously or semi-autonomously controlled vehicle (100), which comprises the following steps: - Determining sensor data using at least one sensor (210) of a vehicle (100), - Processing the sensor data using an artificial neural network (500), wherein the sensor data or data derived from the sensor data are supplied to the artificial neural network (500) as input variables and the artificial neural network (500) maps the input variables to output variables depending on a parameterizable processing chain (410, 420, 430), - Determining at least one value of at least one environmental variable,- Classifying the acquired sensor data depending on the output variables determined by the artificial neural network and depending on the at least one determined value of the at least one environmental variable, and - Executing the control function depending on the classification. Furthermore, the invention provides a control system designed for executing the method, as well as a vehicle with such a control system.