Distributed Edge Training for UAV Fleet Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for updating and retraining machine-learning models used in unmanned aerial vehicle (UAV) delivery fleets are resource-intensive, leading to inaccuracies due to the extensive computing power, network loads, and time required, which worsens as the fleet expands and data collection increases.
Innovation Solution
Utilizing idle UAVs as computing nodes in a distributed computing cluster to perform incremental updates and retraining of machine-learning models, leveraging a computing management module to manage and optimize these processes, allowing for more efficient incorporation of new data and reducing the burden on resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learning models are updated and retrained using centralized computing resources, then model accuracy can be improved, but the system consumes excessive computing power, network bandwidth, and time
Solution Approach 1:
The patent segments the centralized model training process into distributed edge training tasks across multiple UAVs. Each UAV independently trains local model copies using its collected data, dividing the computational workload from a single centralized point to multiple distributed edge devices, thereby reducing centralized computing power consumption and network bandwidth requirements.
Solution Approach 2:
The patent transitions from a single-dimensional centralized training architecture to a multi-dimensional distributed edge training architecture. By adding the spatial dimension of distribution across multiple UAVs and introducing temporal dimension through asynchronous model updates, the system reduces the computational burden on any single resource while maintaining model accuracy.
2Measurement precision
If machine-learning models are retrained frequently with new data, then model accuracy improves, but the time required for training and updating increases
Solution Approach 1:
The patent implements preliminary action by pre-training model copies on each UAV using locally collected data before deployment. This allows the models to be pre-adapted to specific data distributions and scenarios, reducing the time required for subsequent updates and enabling faster response to new data without requiring extensive retraining periods.
Solution Approach 2:
The patent ensures continuity of useful action by implementing continuous incremental learning at the edge. Instead of periodic batch retraining that causes service interruptions, UAVs continuously update their local models as new data arrives, maintaining model accuracy without significant downtime or service disruption.
3Measurement precision
If centralized computing resources are used for model training, then comprehensive data processing is achieved, but the system becomes less adaptable as fleet size expands
Solution Approach 1:
The patent implements universality by creating a multi-functional architecture where each UAV serves dual purposes: performing delivery operations and functioning as an independent training node. This allows the system to scale adaptively as fleet size expands, with each new UAV automatically contributing to model training without requiring proportional increases in centralized computing resources.
Solution Approach 2:
The patent enables self-service by allowing each UAV to independently train and update its own model copies using its locally collected data. This autonomous edge training capability eliminates the bottleneck of centralized processing, allowing the system to automatically adapt to fleet expansion without requiring additional centralized computing infrastructure or manual configuration.
Data Source
AI summary
Systems and methods are provided herein for managing a set of autonomous vehicles (AVs) configured to perform delivery tasks and computing tasks. Computing tasks can be performed such as training a model and/or calculating an incremental update for the model. As additional training data is obtained, a subset of AVs may be managed as a distributed computing cluster and assigned a computing task such as training or calculating an incremental update for the model or any suitable computing task. Corresponding data computed by the subset of AVs of the cluster (e.g., the retrained model, updated model parameters corresponding to the updated model, etc.) may be received and stored or transmitted (e.g., the computing task requestor, to the AVs, etc.) for subsequent use (e.g., for subsequent delivery tasks).