Enhancing intention prediction and interpretability in service robots with LLM and KG.

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Robot Interaction

Background:

• Service robots are increasingly integrated into daily life, requiring accurate human intention recognition.
• Current intention recognition methods face challenges in complex, unstructured environments.
• Effective human intention prediction is crucial for timely and appropriate robot service delivery.

Purpose of the Study:

• To develop an advanced framework for enhancing the intention recognition capabilities of service robots.
• To address the limitations of existing methods in complex and unstructured settings.
• To improve the accuracy, transparency, and interpretability of human intention reasoning in service robots.

Main Methods:

• Proposed the Large language model and Knowledge graph based Intention Intention Recognition Framework (LKIRF).
• Constructed an offline Knowledge Graph (KG) using human motion and environmental data.
• Developed an online reasoning graph via real-time interaction, leveraging Large Language Models (LLMs) for interpretation.

Main Results:

• LKIRF demonstrated improved prediction accuracy across diverse scenarios compared to traditional methods.
• The framework enhanced the transparency of the intention reasoning process.
• Interpretability of the intention reasoning was significantly improved by the proposed approach.

Conclusions:

• The LKIRF effectively enhances service robot intention recognition in complex environments.
• Combining LLMs and KGs offers a powerful approach for human-robot interaction.
• The framework provides more accurate, transparent, and interpretable intention prediction for service robots.