Adaptive Neural Model Scheduling for Autonomous Driving Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex and high-rate model processing in electronic devices, such as those used in autonomous driving, often wastefully utilize resources like processing bandwidth and power, as they perform tasks indiscriminately without adapting to varying conditions like traffic congestion or open roads.
Innovation Solution
Adaptive model processing methods that select between different neural networks based on traffic conditions, adjusting computation rates and transferring computations between processors to optimize resource usage, selecting heavier models for complex conditions like accidents or construction and lighter models for less demanding scenarios like open roads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complex and high-rate model processing is performed indiscriminately, then processing accuracy and speed are maintained, but resource consumption (processing bandwidth and power) increases
Solution Approach 1:
The system dynamically adapts model processing complexity and rate based on real-time driving conditions. When traffic conditions are simple (e.g., open roads), the system reduces model processing rate and complexity. When conditions become complex (e.g., congestion, adverse weather), the system increases processing rate and switches to more complex models, thereby optimizing the balance between processing accuracy and power consumption
Solution Approach 2:
The system changes key processing parameters including model complexity level, processing frame rate, and computational precision based on driving condition assessments. This allows the system to adjust resource consumption dynamically while maintaining adequate processing accuracy for the current operational context
2Reliability
If complex and high-rate model processing is performed indiscriminately, then processing accuracy is maintained, but processing bandwidth utilization increases
Solution Approach 1:
The system dynamically adjusts model processing parameters including computational rate and model complexity based on real-time driving condition assessments. This dynamic adaptation reduces processing bandwidth utilization during simple driving scenarios while maintaining high processing accuracy when complex conditions require enhanced processing capabilities
Solution Approach 2:
The system applies partial processing action by selectively reducing model processing rate and complexity only when full processing capability is not required for safe operation. This allows the system to maintain adequate processing accuracy while avoiding excessive bandwidth utilization during periods of simple driving conditions
3Reliability
If heavier neural networks with more nodes and operations are selected, then model accuracy improves, but computational resources (MACs and FLOPs) increase
Solution Approach 1:
The system changes the complexity parameter of neural network models based on driving conditions. It selects from multiple pre-trained models with varying complexity levels, adjusting model accuracy and corresponding computational requirements dynamically to match the actual operational needs of each driving scenario
Solution Approach 2:
The system dynamically switches between different neural network models of varying complexity. This allows the system to optimize the balance between model accuracy and computational operations by selecting appropriate model complexity levels based on real-time assessments of driving condition complexity
Data Source
AI summary
A method performed by an apparatus is described. The method includes receiving remote data. The method also includes adapting model processing based on the remote data. The method further includes determining a driving decision based on the adapted model processing. In some examples, adapting model processing may include selecting a model, adjusting model scheduling, and/or adjusting a frame computation frequency.