A Unified Model for Multi-Task Drone Routing in Post-Disaster Road Assessment

2025-10-24

AI Summary

Large-scale UAV path planning for post-disaster road assessment faces challenges in multi-task adaptability across diverse problem variants. Method: We propose the first unified deep reinforcement learning (DRL) framework capable of jointly solving eight distinct problem variants. Built upon a Transformer-based encoder-decoder architecture and lightweight adapter-based fine-tuning, it enables synergistic optimization of shared representation learning and task-specific modeling. Contribution/Results: The method achieves efficient inference (1–10 seconds per instance) on thousand-node-scale networks and rapid cross-variant transfer. Compared to single-task DRL baselines, it reduces parameter count and training time by 8× while improving solution quality by 6%–14%; against traditional optimization methods, it yields 24%–82% gains. It significantly enhances generalization capability, deployment flexibility, and real-time decision-making efficacy in emergency response scenarios.

Technology Category

• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Search and Optimization: Learning to Search
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Post-disaster road assessment (PDRA) is essential for emergency response, enabling rapid evaluation of infrastructure conditions and efficient allocation of resources. Although drones provide a flexible and effective tool for PDRA, routing them in large-scale networks remains challenging. Traditional optimization methods scale poorly and demand domain expertise, while existing deep reinforcement learning (DRL) approaches adopt a single-task paradigm, requiring separate models for each problem variant and lacking adaptability to evolving operational needs. This study proposes a unified model (UM) for drone routing that simultaneously addresses eight PDRA variants. By training a single neural network across multiple problem configurations, UM captures shared structural knowledge while adapting to variant-specific constraints through a modern transformer encoder-decoder architecture. A lightweight adapter mechanism further enables efficient finetuning to unseen attributes without retraining, enhancing deployment flexibility in dynamic disaster scenarios. Extensive experiments demonstrate that the UM reduces training time and parameters by a factor of eight compared with training separate models, while consistently outperforming single-task DRL methods by 6--14% and traditional optimization approaches by 24--82% in terms of solution quality (total collected information value). The model achieves real-time solutions (1--10 seconds) across networks of up to 1,000 nodes, with robustness confirmed through sensitivity analyses. Moreover, finetuning experiments show that unseen attributes can be effectively incorporated with minimal cost while retaining high solution quality. The proposed UM advances neural combinatorial optimization for time-critical applications, offering a computationally efficient, high-quality, and adaptable solution for drone-based PDRA.

Problem

• Developing a unified drone routing model for multiple post-disaster road assessment variants
• Overcoming poor scalability of traditional methods and single-task limitations of existing DRL approaches
• Enabling real-time adaptable solutions for dynamic disaster scenarios with minimal retraining

Innovation

• Unified transformer model handles eight routing variants
• Lightweight adapter enables finetuning without full retraining
• Single network reduces parameters while improving solution quality