LLM-Driven Self-Refinement for Embodied Drone Task Planning

2025-08-21

AI Summary

To address the lack of autonomous evaluation and feedback-driven optimization capabilities in industrial-grade embodied drones operating under dynamic, continuous tasks, this paper proposes SRDrone—a novel LLM-driven closed-loop self-optimization framework that integrates logical reasoning with physical execution, overcoming the limitations of conventional single-frame terminal-state assessment. Its key contributions are: (1) a continuous state evaluation mechanism enabling multi-granularity monitoring of task execution; (2) a hierarchical behavior tree online modification model supporting structured reflective learning within constrained policy spaces; and (3) an iterative self-refinement technique leveraging an experience repository to enhance planning robustness over time. Experimental results demonstrate a 44.87% improvement in task success rate over baseline methods, with real-world deployment achieving 96.25% reliability.

Technology Category

• Search and Optimization: Learning to Search
• Intelligent Robots: Learning & Optimization for ROB
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

We introduce SRDrone, a novel system designed for self-refinement task planning in industrial-grade embodied drones. SRDrone incorporates two key technical contributions: First, it employs a continuous state evaluation methodology to robustly and accurately determine task outcomes and provide explanatory feedback. This approach supersedes conventional reliance on single-frame final-state assessment for continuous, dynamic drone operations. Second, SRDrone implements a hierarchical Behavior Tree (BT) modification model. This model integrates multi-level BT plan analysis with a constrained strategy space to enable structured reflective learning from experience. Experimental results demonstrate that SRDrone achieves a 44.87% improvement in Success Rate (SR) over baseline methods. Furthermore, real-world deployment utilizing an experience base optimized through iterative self-refinement attains a 96.25% SR. By embedding adaptive task refinement capabilities within an industrial-grade BT planning framework, SRDrone effectively integrates the general reasoning intelligence of Large Language Models (LLMs) with the stringent physical execution constraints inherent to embodied drones. Code is available at https://github.com/ZXiiiC/SRDrone.

Problem

• Enhancing drone task planning with self-refinement capabilities
• Replacing single-frame assessment with continuous state evaluation
• Integrating hierarchical Behavior Trees with constrained strategy learning

Innovation

• Continuous state evaluation for task outcome assessment
• Hierarchical Behavior Tree modification model integration
• Self-refinement combining LLM reasoning with physical constraints