ORGANA: A Robotic Assistant for Automated Chemistry Experimentation and Characterization

TL;DR

<3-5 sentence high-level summary> Organa presents an assistive robotic system that integrates LLM-based reasoning, perception, and temporal task-and-motion planning with scheduling to automate diverse chemistry experiments in a self-driving-lab–like setting. It translates chemist goals expressed in natural language into parallel, robot-executable plans that coordinate multiple lab devices, including an electrode polishing station for electrochemistry, and generates post-experiment reports. The authors demonstrate solubility screening, recrystallization, pH measurement, and electrochemical characterization of quinones (AQS) with parallel 19-step experiments, achieving meaningful reductions in time and workload in a user study. They also discuss modularity, safety, autonomy, and limitations, and provide open data and code to support further development in flexible lab automation.

Abstract

Chemistry experiments can be resource- and labor-intensive, often requiring manual tasks like polishing electrodes in electrochemistry. Traditional lab automation infrastructure faces challenges adapting to new experiments. To address this, we introduce ORGANA, an assistive robotic system that automates diverse chemistry experiments using decision-making and perception tools. It makes decisions with chemists in the loop to control robots and lab devices. ORGANA interacts with chemists using Large Language Models (LLMs) to derive experiment goals, handle disambiguation, and provide experiment logs. ORGANA plans and executes complex tasks with visual feedback, while supporting scheduling and parallel task execution. We demonstrate ORGANA's capabilities in solubility, pH measurement, recrystallization, and electrochemistry experiments. In electrochemistry, it executes a 19-step plan in parallel to characterize quinone derivatives for flow batteries. Our user study shows ORGANA reduces frustration and physical demand by over 50%, with users saving an average of 80.3% of their time when using it.

Figures (21)

• Figure 1: Robot setup with Organa's overall schema.Organa provides seamless interaction between SDLs and chemists for diverse chemistry experiments. A key strength of Organa is that it perceives surrounding objects and keeps track of progress on a chemistry task in order to make an informed decision about next steps that are in line with user goals. Organa optimizes SDL efficiency through parallel experiment execution, providing timely feedback via reports and analysis, keeping users well-informed and involved in high-level decision-making. More information about Organa can be found at https://ac-rad.github.io/organa/, including code and a video demonstration.
• Figure 2: Instances of Organa conducting various chemistry experiments: (A) solubility, (B) recrystallization, and (C) pH testing. Images showcase the robot executing actions in each setup, as well as the final results.
• Figure 3: The electrochemistry setup and experiment workflow. Initially, users communicate their intention to Organa via a text or speech interface; objects and their poses are perceived. Subsequently, the user interacts with Organa to establish object functionalities. Eventually, Organa plans the robot actions for parallel execution. On top, it shows the experiment setup. At the bottom, it displays the results of visual perception and snapshots of the robot and other hardware executing the actions in parallel.
• Figure 4: The electrochemistry results executed by Organa. On the left, a Pourbaix diagram is shown for a single Organa experimental run with estimated distributions for p$K_{\mathrm a1}$ and p$K_{\mathrm{a2}}$. The maximum likelihood estimation (MLE) for p$K_{\mathrm{a1}}$ is 7.86. The top right plot is the cyclic voltammetry curve at pH$=9$. In the bottom right, the estimated slope distribution for the first region is shown, with an MLE of -61.8 mV/pH unit. The distributions for $pK_{\mathrm{a1}}$, $pK_{\mathrm{a2}}$, and the slope are marginal distributions for individual parameters. We report the MLE from the full combined posterior distribution over all model parameters in \ref{['note:appendix:report']} , which may be different from the maximum value in the marginal distribution for any given parameter.
• Figure 5: Gantt chart of the electrochemistry experiment with sequential and parallel execution. The execution times are written in the legend. For sequential execution as shown in (a), only one action is executed at a given time and it takes 1,346 seconds (22.43 min) to complete a single test of a buffer solution. For parallel execution shown in (b), the total execution decreases to 1,071 seconds (17.85 min) since multiple actions associated with different agents can be executed at the same time. Note: pump cannot transfer solution to pH beaker at $\sim$500 sec in the parallel execution experiment due to ongoing redox potential measurement. Boxes stacked on top of each other with the same color and pattern mean that the agents are involved in performing the same action jointly.