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

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.