Electromagnetic metamaterial agent

Area of Science:

• Metamaterials and intelligent systems
• Artificial intelligence in wave control

Background:

• Metamaterials have advanced from passive to self-adaptive devices.
• Deep learning currently aids metamaterial design but has limitations in autonomous functionality.

Purpose of the Study:

• To propose and prototype a paradigm shift towards a metamaterial agent (metaAgent) with cognitive capabilities.
• To enable autonomous planning and execution of complex, long-horizon tasks by metamaterials.

Main Methods:

• Developed a metaAgent leveraging foundation models for natural language reasoning and task planning.
• Implemented a multi-agent discussion mechanism for sensing, planning, grounding, and coding.
• Integrated environmental feedback and human requests for real-world task execution.

Main Results:

• The metaAgent prototype demonstrated autonomous planning and execution of electromagnetic (EM) field manipulations.
• Successfully integrated EM tasks with robot control in an ambient-assisted living context.
• Showcased metaAgent's ability to master foundational EM skills and learn from human feedback.

Conclusions:

• The metaAgent represents a significant advancement, moving beyond predefined functionalities to autonomous, cognitive operation.
• This cognitive metamaterial agent can autonomously manage complex EM interactions and adapt to dynamic environments.
• Future applications include advanced wireless communications, sensing, and human-robot interaction.