# Agents for self-driving laboratories applied to quantum computing

**Authors:** Shuxiang Cao, Zijian Zhang, Mohammed Alghadeer, Simone D Fasciati, Michele Piscitelli, Mustafa Bakr, Peter Leek, Alán Aspuru-Guzik  
**arXiv:** [2412.07978](https://arxiv.org/abs/2412.07978)  
**Date:** 2024-12-10  
**Summary page:** https://www.summarizepaper.com/en/paper/2412.07978v1/

## Key points

- Fully automated self-driving laboratories enable high-throughput and large-scale research endeavors
- Utilize intelligent agents to streamline experiments and manage laboratory knowledge
- Introduction of the k-agents framework to integrate complex laboratory knowledge effectively
- Large language model-based agents organize laboratory knowledge and automate experiments
- Collaborate with execution agents for multi-step protocols and thorough result analyses
- Utilize analyzed data for effective state transitions and closed-loop feedback control
- Successfully deployed in calibrating and operating a superconducting quantum processor autonomously
- Generates and characterizes entangled quantum states at a level comparable to human scientists

## Summary

In the realm of scientific discovery, fully automated self-driving laboratories have emerged as a promising avenue for enabling high-throughput and large-scale research endeavors. These labs alleviate the burden of repetitive tasks by utilizing intelligent agents to streamline experiments and manage laboratory knowledge. The efficacy of automation relies on seamlessly integrating complex laboratory knowledge, which is often unstructured and challenging to incorporate into existing AI systems. To address this challenge, a groundbreaking k-agents framework has been introduced in this paper. This innovative framework empowers experimentalists by organizing laboratory knowledge and automating experiments through the use of intelligent agents. At its core are large language model-based agents that serve as repositories for diverse laboratory operations and methodologies. These agents work with execution agents to break down multi-step protocols into manageable state machines and collaborate with other specialized agents to execute each step seamlessly while also conducting thorough analyses of results. One notable feature of this framework is its ability to utilize analyzed data for effective state transitions, facilitating closed-loop feedback control within the experimental process. To demonstrate its capabilities, the k-agents were deployed in calibrating and operating a superconducting quantum processor autonomously. Over an extended period, these agents successfully planned and executed intricate experiments with unparalleled precision, generating and characterizing entangled quantum states at a level comparable to human scientists. The implementation of this knowledge-based agent system represents a significant leap forward in managing laboratory knowledge effectively and accelerating scientific discovery across various domains. By bridging the gap between complex laboratory procedures and cutting-edge AI technologies, the k-agents framework holds immense promise for revolutionizing research practices and unlocking new frontiers in scientific exploration.