FCRF: Flexible Constructivism Reflection for Long-Horizon Robotic Task Planning with Large Language Models

TL;DR

This paper introduces FCRF, a new framework enabling large language models to adaptively reflect on and learn from failures in complex robotic tasks, significantly enhancing autonomous error correction and task success rates.

Contribution

FCRF presents a flexible, human-inspired self-reflection mechanism for LLMs that adaptively integrates experience and failure lessons for improved robotic task planning.

Findings

FCRF outperforms baseline methods in simulation and real-world tasks.

FCRF enhances the flexibility and effectiveness of self-reflection in LLMs.

Experimental results show improved success rates in long-horizon tasks.

Abstract

Autonomous error correction is critical for domestic robots to achieve reliable execution of complex long-horizon tasks. Prior work has explored self-reflection in Large Language Models (LLMs) for task planning error correction; however, existing methods are constrained by inflexible self-reflection mechanisms that limit their effectiveness. Motivated by these limitations and inspired by human cognitive adaptation, we propose the Flexible Constructivism Reflection Framework (FCRF), a novel Mentor-Actor architecture that enables LLMs to perform flexible self-reflection based on task difficulty, while constructively integrating historical valuable experience with failure lessons. We evaluated FCRF on diverse domestic tasks through simulation in AlfWorld and physical deployment in the real-world environment. Experimental results demonstrate that FCRF significantly improves overall performance and self-reflection flexibility in complex long-horizon robotic tasks.