An Agentic AI Framework with Large Language Models and Chain-of-Thought for UAV-Assisted Logistics Scheduling with Mobile Edge Computing
Summary: arXiv:2605.13221v1 Announce Type: new
Abstract: In the realm of cloud manufacturing, the integration of unmanned aerial vehicles (UAVs) represents a significant advancement in logistics and computational task management. This new study presents a comprehensive framework for optimizing the joint operations of UAV-assisted logistics and mobile edge computing (MEC), addressing the complexities of hybrid scheduling problems.
Introduction
The increasing demand for efficiency in manufacturing processes has led to the innovative use of UAVs in logistics. These aerial vehicles not only facilitate product collection but also enhance computational capabilities through mobile edge computing. The synthesis of these two functions creates a hybrid scheduling challenge that necessitates advanced problem-solving approaches.
The Hybrid Scheduling Problem
The hybrid scheduling problem involves coordinating UAV operations for product collection while simultaneously managing computational tasks generated by industrial sensors. Key challenges include:
- Routing Decisions: The routing of UAVs directly influences when and how computational tasks can be offloaded to the cloud, affecting overall logistics efficiency.
- Energy Budget: UAVs must manage their energy consumption effectively to maintain operational capabilities throughout their service windows.
- Resource Allocation: Availability of onboard computing and communication resources is vital for executing computational tasks within specified deadlines.
The Proposed Framework
To tackle these challenges, the authors propose an agentic-AI-assisted optimization framework comprising two main components:
- Agentic AI Development: This component integrates large language models, retrieval-augmented generation, and chain-of-thought reasoning. It translates user input into a clear mathematical formulation for the hybrid scheduling problem, making it more interpretable and manageable.
- Hierarchical Deep Reinforcement Learning: Utilizing proximal policy optimization (PPO), this approach features a two-layer structure. The upper layer focuses on UAV routing, while the lower layer is dedicated to optimizing task execution and resource allocation on a per-slot basis.
Simulation Results
Extensive simulations were conducted to evaluate the effectiveness of the proposed framework. The results indicate:
- The agentic AI produced consistently accurate formulations of the scheduling problem.
- The hierarchical PPO achieved a remarkable 99.6% success rate in full product collection across the last 500 episodes.
- There was a 100% satisfaction rate regarding task deadlines, showcasing the reliability of the framework.
- Performance stability was notably superior compared to the traditional advantage actor-critic approach.
Conclusion
This innovative framework demonstrates the potential of combining agentic AI with advanced reinforcement learning techniques to optimize UAV-assisted logistics and mobile edge computing. As industries continue to evolve, such solutions will be crucial in enhancing operational efficiencies and meeting the demands of modern manufacturing environments.
Future research may explore further integration of AI methodologies and the extension of this framework to other logistics and computational scenarios, paving the way for smarter, more responsive manufacturing systems.
Related AI Insights
- Transferable User Preferences for Human-Aligned AI Decisions
- Realistic User Personas for Robust LLM Agent Evaluation
- Strikingness-Aware Evaluation for Temporal Knowledge Graphs
- BEHAVE: Hybrid AI for Real-Time Human Group Dynamics
- Multimodal HMMs for Persistent Emotional State Tracking
- KITE: AI Tutoring for Algorithm Tracing & Problem-Solving
- Bot-Mod: Advanced Multi-Turn Dialogue for Intent Detection
- PROMETHEUS: Automating Deep Causal Research with AI Models
- Sustaining AI Safety: Control Limits & Structural Needs
- Why LLMs Lose Context in Multi-Turn Conversations
