Edge Server Selection for In-Vehicle Media via Deep Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in efficiently downloading large media files due to limited time within range of individual edge servers, necessitating improved methods for partitioning media content across multiple edge servers to minimize latency.
Innovation Solution
A system utilizing deep reinforcement learning and Graph Neural Networks to determine optimal content partitioning across edge servers based on vehicle trajectories, edge server connections, and network bandwidth, ensuring efficient media content delivery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If media content is downloaded from a single edge server, then the download process is simple, but the vehicle cannot complete the download within the limited time it is within range of that edge server
Solution Approach 1:
The media content is divided into multiple partitions, with each partition downloaded from a different edge server. The system determines optimal partitioning based on the vehicle's trajectory and the coverage areas of multiple edge servers, allowing the download to be completed within the vehicle's limited range of each server while minimizing total download time.
2Productivity
If media content is partitioned across multiple edge servers, then the download can be completed within limited range time, but the system complexity increases due to need for trajectory prediction and server coordination
Solution Approach 1:
The system performs preliminary actions by predicting the vehicle's future trajectory and pre-determining the optimal partitioning of media content across edge servers before the download begins. This advance planning allows the system to coordinate multiple edge servers efficiently without real-time complexity, as the partitioning strategy is established in advance based on predicted vehicle movement.
3Adaptability or versatility
If the vehicle moves through multiple edge server coverage areas, then more edge servers are available for download, but determining optimal connections and partitioning becomes more difficult
Solution Approach 1:
The system uses reinforcement learning with a reward function that provides feedback on the quality of partitioning decisions. The reward function considers factors such as download time, trajectory accuracy, and edge server coverage, allowing the system to learn optimal connection strategies through iterative training. This feedback mechanism simplifies the decision-making process by converting complex multi-server coordination into a learnable optimization problem.
Data Source
AI summary
A method is provided including receiving a planned route of a vehicle and a request to download content from a cloud server, the planned route traveling through an area covered by a plurality of edge servers, determining a state comprising possible connections between the vehicle and each of the plurality of edge servers at a plurality of time steps during the planned route, inputting the state to a trained model, the model being trained to output an action comprising a partition of the content across the plurality of edge servers that minimizes latency of transmission of the content from the cloud server to the vehicle via the plurality of edge servers, based on the state, and partitioning the content across the plurality of edge servers based on the action out by the trained mode.