Neural Network Association Costs for Autonomous Vehicle Perception
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Designing optimal association functions for multi-sensor systems, such as those used in autonomous vehicles, is challenging due to the need for manual iteration in determining distance metrics and weights, which is time-consuming and prone to change with new data or algorithm updates.
Innovation Solution
Employing machine learning models, specifically deep neural networks like Multi-Layer Perceptron (MLP) models, to automatically determine association costs between sensor measurements and predicted states, eliminating the need for manual design of association functions and allowing for updates based on new data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If manual iteration is used to determine distance metrics and weights for association functions, then the system can be designed with conventional methods, but the process is time-consuming and requires frequent manual updates
Solution Approach 1:
The patent replaces the manual mechanical process of designing association functions with an automated machine learning system. The ML model automatically learns optimal association functions from sensor data and predicted states, eliminating the need for manual iteration and significantly reducing development time while improving ease of manufacturing the perception system
Solution Approach 2:
The machine learning model performs self-learning and self-optimization to determine association functions. The system automatically updates its own parameters and metrics based on training data, enabling self-service capability that eliminates dependency on manual intervention for designing and updating association functions
2Device complexity
If conventional association functions are used, then the design process is simpler, but the system lacks adaptability when dependencies change
Solution Approach 1:
The patent implements dynamic association functions through machine learning models that can adapt and update automatically when dependencies change. The ML model learns from training data and can retrain when new sensor data or predicted states are available, providing dynamic adaptability without requiring complete redesign of the association functions
Solution Approach 2:
The system changes parameters automatically through machine learning by adjusting weights, distance metrics, and association thresholds based on learned patterns from training data. This allows the association functions to adapt to changing conditions by modifying their parameters rather than requiring structural changes to the system architecture
3Extent of automation
If machine learning models are used to determine association costs, then adaptability and automation are improved, but the computational complexity increases
Solution Approach 1:
The patent applies preliminary action by training the machine learning model offline before deployment. The association functions are determined in advance through training on historical sensor data and predicted states, allowing the model to be deployed with pre-learned parameters that reduce real-time computational complexity while maintaining high automation
Data Source
AI summary
In various examples, systems and methods are disclosed relating to determining associations between objects represented in sensor data and predicted states of the objects in multi-sensor systems such as autonomous or semi-autonomous vehicle perception systems. Systems and methods are disclosed that employ neural network models, such as multi-layer perceptron (MLP) models or other deep neural network (DNN) models, in learning association costs between sensor measurements and predicted states of objects. During training, the systems and methods can generate data for updating parameters of the neural network models such that, during deployment, the neural network models can receive sensor data and predicted states, and provide corresponding association costs.