Selective Model Execution in Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in efficiently managing multiple machine learning models for real-time operations due to substantial computational resource usage and time constraints, leading to delays and errors in state transitions.
Innovation Solution
Selective model execution in autonomous vehicles involves identifying neighboring state spaces associated with corresponding machine learning models and executing those models to reduce computational load and improve transition efficiency, by prohibiting the execution of non-neighboring models and ensuring preconditions are met before state transitions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple machine learning models are executed concurrently to handle real-time autonomous vehicle operations, then the system can maintain comprehensive monitoring and control across different state spaces, but the computational resource usage increases substantially and causes delays in state transitions
Solution Approach 1:
The patent segments the machine learning model execution by dividing state spaces into current and neighboring categories. Only models corresponding to neighboring state spaces are executed concurrently, while models for non-neighboring states are prohibited from execution. This segmentation reduces the number of concurrently executed models while maintaining system reliability through selective monitoring of relevant state transitions.
2Reliability
If multiple machine learning models are executed concurrently to ensure comprehensive state space coverage, then the system can prevent errors and delays in state transitions, but the computational complexity and resource management difficulty increase
Solution Approach 1:
The patent applies preliminary action by pre-identifying which machine learning models correspond to neighboring state spaces before execution begins. The system determines in advance which models should be executed concurrently based on the current state space, eliminating the need for complex real-time decisions about model execution. This preliminary classification simplifies resource management while ensuring comprehensive coverage of relevant state transitions.
Data Source
AI summary
Selective model execution in an autonomous vehicle, including: identifying, based on a first state space associated with a first machine learning model, one or more second state spaces neighboring the first state space, wherein the one or more second state spaces are each associated with a corresponding second machine learning model of one or more second machine learning models; and executing the one or more second machine learning models.