Distributed Neural Network Processing for Autonomous Vehicle Sensor Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face limitations in processing sensor data due to constrained onboard resources, leading to latency issues and scalability problems in conventional systems that partition processing between the vehicle and the cloud, especially in areas with poor network connectivity.
Innovation Solution
A distributed neural network processing architecture that leverages vehicle proximity and local compute resources, using 5G mmWave communication for fast data exchange between vehicles to split neural network processing across multiple nodes, reducing reliance on cloud offloading and enhancing data processing speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If sensor data processing is performed on onboard vehicle systems, then processing speed and latency are improved, but system complexity and resource requirements increase beyond what single vehicles can support
Solution Approach 1:
The patent segments the neural network processing workload across multiple vehicle nodes in a distributed manner. Each vehicle contributes its compute resources to process portions of the sensor data, breaking down the complex processing task into manageable segments that can be handled by individual vehicles with limited resources.
Solution Approach 2:
The patent merges compute resources from multiple proximate vehicles to create a collective processing power that exceeds what any single vehicle possesses. By combining the computational capabilities of nearby vehicles, the system achieves high-performance processing while maintaining low latency through local collaboration.
2Productivity
If more compute resources are allocated to sensor data processing, then processing accuracy and speed are improved, but onboard platform resources become insufficient
Solution Approach 1:
The patent transitions from a single-vehicle resource model to a multi-vehicle distributed resource model. By adding the spatial dimension of multiple vehicles contributing resources, the system achieves abundant compute power without requiring each individual vehicle to have sufficient onboard resources.
Solution Approach 2:
Each vehicle in the network serves multiple functions: it processes its own sensor data, contributes compute resources to other vehicles, and participates in collective processing tasks. This multi-functionality maximizes the utility of available resources across the entire vehicle network.
3Quantity of substance
If cloud offloading is used for processing sensor data, then resource constraints are relieved, but network connectivity requirements increase and latency is worsened in areas with poor connectivity
Solution Approach 1:
The patent introduces proximate vehicles as intermediary processing nodes between the sensor data source and the final processing destination. Instead of directly offloading to distant cloud infrastructure, data is processed through local vehicle intermediaries that are always available, ensuring reliable processing regardless of cloud connectivity status.
Solution Approach 2:
The system performs processing actions locally at proximate vehicles before data would need to be transmitted to the cloud. By completing processing tasks in advance using available local resources, the system eliminates the need for reliable cloud connectivity and reduces latency.
Data Source
AI summary
Various systems and methods for implementing distribution of a neural network workload are described herein. A discovery message is encoded that includes a latency requirement and requested resources for a workload of a neural network. A discovery response, from a proximate resource and in response to the discovery message, is decoded and includes available resources of the proximate resource available for the workload based on the requested resources for the workload. The proximate resource is selected to execute the workload based on the available resources of the proximate resource. In response to the discovery response, an offload request is encoded that includes a description of the workload. The description of the workload identifies the node to execute at the proximate resource. In response to the offload request, an input is provided to a ADAS system based on the result.