Agentic LLM-based robotic systems for real-world applications: a review on their agenticness and ethics

Area of Science:

• Robotics and Artificial Intelligence
• Human-Robot Interaction
• Machine Learning Ethics

Background:

• Agentic AI systems operate autonomously, making decisions and taking actions with minimal human input.
• Large Language Models (LLMs) offer advanced reasoning and knowledge capabilities, enabling new possibilities for robotic autonomy.
• Integrating LLMs into robotic systems is a rapidly developing field with significant real-world applications.

Purpose of the Study:

• To provide a comprehensive survey of LLM-based robotic systems exhibiting agentic behaviors in real-world applications.
• To introduce a novel classification system for evaluating the 'agenticness' of LLM-driven robots.
• To establish an ethical and safety evaluation framework for these advanced robotic systems.

Main Methods:

• Systematic categorization of LLM-based robotic systems across key application domains (navigation, manipulation, multi-agent, multi-task).
• Development of a novel agenticness classification based on autonomy, goal-directedness, adaptability, and decision-making.
• Creation of an evaluation framework for ethical, safety, and transparency principles (bias, fairness, robustness, safety, oversight, explainability, auditability, compliance).

Main Results:

• Identified and categorized diverse LLM-driven robotic systems demonstrating validated agentic behaviors.
• Proposed a new metric to quantify the level of agenticness in LLM-powered robots.
• Highlighted key gaps, tensions, and trade-offs in current approaches concerning capabilities and ethical safeguards.

Conclusions:

• LLM integration is advancing robotic agentic capabilities significantly.
• A robust framework is crucial for evaluating and ensuring the ethical and safe deployment of these systems.
• Continued research is essential to align increasingly capable LLM-powered robots with societal norms and values.