Multi-Robot Systems for Laboratory Automation

This looks like it’s a science fiction animation, but it’s actually a real-life video of our latest generation robotic systems from Multiply Labs (watch the end of the video to see the proof)! Seriously though, extreme parallelism is one of the (many) areas where robotic systems reach levels of throughput and efficiency that are simply impossible to match manually. As cell, gene, and mRNA therapies move from clinical trials into the commercial stage, every minute and every square foot in your clean-room becomes critical. Manual hand-offs or even isolated robotic arms become automatic throughput bottlenecks - there simply isn’t enough room (or people) to scale at the pace patients need. That’s why we went with 4 fully synchronized arms. By mounting 2 overhead and 2 on rails, we’ve more than tripled the output in the same footprint and slashed cycle times. Hot-swap end-effectors keep syringe transfers, bag shuttling, and vial handling flowing without pause - no human in the loop, no downtime for changeovers. Stay tuned for next week’s #LabNotes, where we’ll take a deeper dive into multi-robotic arm parallelism!

Imagine running 3,600 synthesis experiments in a single day and learning something fundamental from each one. That's the power of Self-Driving Labs (SDLs). When I bring up SDLs with colleagues in biotech, the concept clicks instantly. The field already leans on robotic assays, high-throughput screening, and informatics pipelines. So it feels like a natural extension to integrate AI/ML for closed-loop optimization. But when I speak with peers in chemistry and materials science, I often encounter more hesitation. The idea of handing over experimental decisions to machines can feel at odds with a culture shaped by first-principles reasoning, serendipity, and manual iteration. That's why I was so intrigued when I met Prof. Milad Abolhasani at the NIST AI for Materials Science (AIMS) workshop last month. His lab’s work offers a compelling example of the SDL vision for materials discovery, bridging the gap between possibility and practice. From catalysis optimization to quantum materials, Abolhasani’s team is building SDL platforms that not only automate experiments, but also accelerate understanding. A recent standout is Rainbow: a multi-robot, AI-powered system that autonomously explores the synthesis of metal halide perovskite nanocrystals. By combining miniaturized batch reactors, real-time spectral feedback, and multi-objective Bayesian optimization, Rainbow executed thousands of synthesis trials in a single day: mapping Pareto fronts, uncovering structure–property relationships, and scaling up the best results. Maybe now is the time to imagine what SDLs could mean for materials innovation in your organization. 📄 Autonomous multi-robot synthesis and optimization of metal halide perovskite nanocrystals, Nature Communications, August 22, 2025 🔗 https://lnkd.in/euz72pXe

A new study out of China claims AI agents can coordinate to run biology experiments without human input 🤖. At least, to some extent... 📑 Just read the paper “BioMARS: A Multi-Agent Robotic System for Autonomous Biological Experiments” (link in the comments). The paper was authored by researchers at the University of Science and Technology of China (USTC) in Suzhou and Hefei. It describes a system that combines large language models, robotic control, and multimodal perception to automate biological experiments end-to-end. The architecture is split across three AI agents: ⚙️ Biologist Agent: uses retrieval-augmented generation to design protocols based on literature and lab databases. ⚙️ Technician Agent: converts those protocols into structured instructions compatible with specific lab hardware (e.g. pipetting, incubation). ⚙️ Inspector Agent: monitors execution using visual and sensor data to detect anomalies (e.g. incorrect fluid volumes, contamination). The system was tested on standard lab tasks like cell passaging and stem cell differentiation. Authors claim it performed comparably to humans in terms of viability and consistency, and in some cases optimized outcomes beyond manual protocols. There’s also a browser interface for human oversight and modular integration with lab equipment, allowing for both autonomy and human-in-the-loop operation (see image below). Technically, it’s an example of structured LLM orchestration with physical task grounding, multimodal feedback loops, and real-world robotic execution. I am not sure about edge cases and stability of the system, but the article is worth a read if you're interested in the practical application of LLMs in lab automation or scientific workflows. Image reproduced from the paper (link in the comments).