Azala

Localization, inspection, and reasoning (LIRA) module for autonomous workflows in self-driving laboratories

Gorde:
Argitaratua izan da:Communications Chemistry vol. 8, no. 1 (2025), p. 384-396
Egile nagusia: Zhou, Zhengxue
Beste egile batzuk: Veeramani, Satheeshkumar, Munguia-Galeano, Francisco, Fakhruldeen, Hatem, Cooper, Andrew I.
Argitaratua:
Nature Publishing Group
Gaiak:
Failure
Adaptability
Error correction
Nuclear magnetic resonance > NMR
Workflow
Edge computing
Laboratories
Robots
Modules
Automation
Localization
Decision making
Error detection
Robotics
Inspection
Reasoning
Artificial intelligence
Real time
Client server systems
Large language models
Human error
Semantics
Sarrera elektronikoa:Citation/Abstract
Full Text
Full Text - PDF
Etiketak: Etiketa erantsi
Etiketarik gabe, Izan zaitez lehena erregistro honi etiketa jartzen!
Laburpena:Self-driving labs (SDLs) combine robotic automation with artificial intelligence (AI) to allow autonomous, high-throughput experimentation. However, robot manipulation in most SDL workflows operates in an open-loop manner, lacking real-time error detection and error correction. This can reduce reliability and overall efficiency. Here, we introduce LIRA (Localization, Inspection, and Reasoning), which is an edge computing module that enhances robotic decision-making through vision-language models (VLMs). LIRA enables precise localization, automated error inspection, and reasoning, thus allowing robots to adapt dynamically to variations from the expected workflow state. Integrated within a client-server framework, LIRA supports remote vision inspection and seamless multi-platform communication, improving workflow flexibility. Through extensive testing, LIRA achieves high localization accuracy, a tenfold reduction in localization time, and real-time inspection across diverse tasks, increasing the efficiency and robustness of autonomous workflows considerably. As an open-source solution, LIRA facilitates AI-driven automation in SDLs, advancing autonomous, intelligent, and resilient laboratory environments. Longer term, this will accelerate scientific discoveries through more seamless human-machine collaboration.Robotic automation in self-driving laboratories often involves workflows that are run in an open-loop manner, lacking real-time error detection and error correction. Here, the authors introduce a computing module that enhances robotic decision-making through vision language models, facilitating dynamic adaption to variations in the expected workflow.
ISSN:2399-3669
DOI:10.1038/s42004-025-01770-1
Baliabidea:Materials Science Database