Decentralized Multivehicle Delivery With RL-GNN Route Pooling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional transport techniques rely on centralized decision-making authorities, leading to inefficiencies and limited ability to address delays, shortages, and demand spikes, and fail to effectively pool travel requests for efficient service provision.
Innovation Solution
A decentralized on-demand multivehicle delivery system using a reinforcement learning-based graph neural network (RL-GNN) to generate node graphs and optimize vehicle routes, maximizing service provision and minimizing delays by pooling travel requests across a network of service vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a centralized decision-making authority is used to manage vehicle routes and assignments, then coordination and control are simplified, but the system's ability to respond to demand spikes and delays is limited
Solution Approach 1:
The patent divides the centralized routing problem into smaller sub-problems by creating separate graphs for different vehicles and time periods. The overall route optimization is segmented into manageable chunks that can be processed independently yet coordinated through the graph neural network framework.
Solution Approach 2:
The system dynamically adapts to changing conditions by using reinforcement learning to adjust vehicle assignments and routes in real-time based on demand fluctuations, delays, and other operational factors. The graph neural network continuously learns from new data to optimize decisions under varying conditions.
2Ease of manufacture
If travel requests are processed individually without pooling, then individual request handling is simple, but service efficiency and resource utilization are reduced
Solution Approach 1:
The patent merges multiple individual travel requests into a unified graph structure that captures relationships between requests, vehicles, and locations. This allows the system to process requests collectively and identify optimization opportunities that would be invisible when handling requests individually.
Solution Approach 2:
The graph neural network framework serves multiple functions simultaneously: it processes individual request constraints, optimizes vehicle assignments, coordinates routes across multiple vehicles, and adapts to changing conditions. This multi-functional approach maintains simplicity while achieving high service efficiency.
3Reliability
If a comprehensive request-vehicle graph is created for all vehicles, then complete system optimization is achieved, but computational complexity and processing time increase
Solution Approach 1:
The patent segments the comprehensive request-vehicle graph into smaller sub-graphs, each representing a specific vehicle's operational context. This segmentation reduces the computational burden while preserving the essential relationships needed for optimization through the graph neural network.
4Speed
If real-time data processing is implemented to handle fluctuating demand, then service responsiveness is improved, but computational resources and processing time are increased
Solution Approach 1:
The system dynamically adjusts its processing intensity and scope based on real-time conditions, focusing computational resources on the most critical decisions and vehicles. The graph neural network efficiently processes only the necessary data subsets required for current optimization needs.
Data Source
AI summary
A computing device and a method to generate a decentralized delivery scheme for multivehicle in an interested area may comprise the following steps. First, the computing device may receive map data, travel request data, and service vehicle data. Second, the computing device may determine one or more constraints based on the travel request data. Third, the computing device may abstract the travel request data, the service vehicle data, and the map data into a request-vehicle graph comprising nodes and edges. Fourth, the computing device may trim the request-vehicle graph into partial request-vehicle graphs for each service vehicle. Fifth, the computing device may encode the partial request-vehicle graphs through a graph neural network (GNN). Sixth, the computing device may train the GNN to predict actions for each service vehicle. Finally, the computing device may instruct the service vehicles to operate based on the actions.