Poster sessions will take place on August 12 and 13 at the following venue.
Poster sessions will take place on August 12 and 13 at the following venue.
Posters will be on view throughout the day, as well as during a dedicated poster session with refreshments and snacks.
We introduce a semi-automated workflow and open-source software pyGecko for precise analysis of high-throughput reaction screening without external calibration. This accelerates chemical discoveries and paves the way for downstream machine learning applications.
This work focuses on designing tools to automate microgel synthesis to identify top performing inflammation-targeting microgel formulations for drug delivery.
This study presents an AI-guided and automation-integrated workflow for developing IrRu-based OER catalysts. Multi-Objective AI model and automated synthesis and evaluation systems were employed to optimize catalyst compositions based on activity, stability, and cost.
Rapid protein solution viscosity estimation as a function of concentration. The presented machine learning model combines graph and language components to incorporate both structural and sequential protein features.
We present a modular platform for the development of graph-based EI-MS spectra prediction models, supporting the entire ML lifecycle.
Instead of relying on human intuition to define experimental boundaries, our system learns them on its own—making autonomous labs more flexible, scalable, and truly self-directed.
We present an AI-powered platform that automates the nanoparticle washing process by integrating robotic arms, computer vision, and large language models (LLMs). This innovative system enhances reproducibility and efficiency in chemical experiments by adapting dynamically to visual and cognitive changes in the experimental environment.
DIGIBAT provides automated workflows covering materials synthesis, electrode preparation, coin cell assembly, and electrochemical testing. By leveraging robotics and automated data collection, the facility reduces experimental time while enhancing reproducibility towards scalability. Our goal is to accelerate energy research, shorten the materials discovery timeline, and enable AI-powered innovation in battery and Power-to-X development.
We use several different spectroscopies to monitor chemical reactions in real time. Evaluating all spectra jointly allows us to identify unknown (intermediate) products and control multistep reactions in detail.
This study proposes an inverted configuration where the mortar is mounted on the robotic arm, enabling a single robot to execute continuous grinding and cleaning. In the next phase, integrating this system with characterization tools will move toward a fully autonomous loop, encompassing powder processing, synthesis, and subsequent evaluation.
We developed a machine learning approach that accelerates the discovery of efficient catalysts by using quantum-inspired similarity to reduce computational cost. This method successfully identified a promising new material for clean energy applications.
We developed autonomous robotic experimental setup that can find out the suitable parameter conditions for controlling the coffee ring effect. The system was coupled with AI-driven interface, NIMO (NIMS Orchestration System), to facilitate autonomous experimentation. Additionally, the PDC (Phase Diagram Construction) method was employed to efficiently explore the boundary condition of coffee ring formation.
EvoMPFs are optimized molecular representations for machine learning, derived using an evolutionary algorithm that identifies the most relevant structures in a set of molecules. This enables researchers to make accurate chemical predictions while revealing meaningful patterns in their data.
Current limitations in laboratory robot control systems hinder the automation of complex tasks like powder mixing in fuel cell research, where manual experimentation remains prevalent. This study explores the application of imitation learning to the complex task of powder grinding, aiming to demonstrate its potential for advancing laboratory automation by comparing its performance to existing methods.
This work presents a novel, cost-effective control platform for automated powder dispensing, integrating an analytical balance, a powder dispenser, and a low-cost motor. The system enables robust, sub-milligram-level dispensing performance across various powder types, including free-flowing and highly cohesive materials.
Due to reporting bias, the yield prediction of chemical reactions, still suffers from the lack of negative data reported in literature. Positivity Is All You Need (PAYN) is Python-based framework that tackles this challenge by using positive-unlabeled learning.
We introduce CombineNet, a general-purpose neural network potential incorporated with adaptive dispersion and electrostatic corrections. The model describes long-range intermolecular interactions with physical asymptotic behaviour and corrects the potential energy prediction inherited from density-functional theory reference methods.
A novel human-informed machine learning workflow combining pre-trained models with state-of-the-art computer vision demonstrates efficiency and effectiveness at image-based cell culture analysis, indicating its suitability for integration into an autonomous biological self-driving laboratory platform.
ATLAS is an automated high-throughput facility at Imperial College London, enabling rapid synthesis, sample preparation, and characterisation for materials and molecular discovery. By integrating robotic workflows with fast, data-rich analysis, ATLAS helps researchers in academia and industry accelerate innovation across chemistry and materials science.
Robot arm performs organ-on-a-chip cultures with a speed and precision surpassing human lab-workers.
We explore the benefits of employing autonomous experimentation for rapid and intelligent navigation of the complex and unintuitive alloy design space. Our approach builds a foundation for autonomous experimentation by leveraging both design principles and hardware from industrial automation to enable high-throughput sample printing, transportation, grinding, polishing, cleaning, microscopy, and hardness testing to capture the lifecycle of alloys.
We present progress towards a machine learning approach that predicts the future state of a battery given the current state and a charging protocol; adapting the charging protocols to the battery state is known to improve battery health. Our approach is specifically designed to be integrated into self-driving labs for accelerated materials discovery, and leverages representations of battery degradation extracted from EIS/transport measurements to make predictions.
We deployed and evaluated a teleoperation system that allows remote human intervention in self-driving laboratories to recover from unexpected errors and maintain smooth operation. By using a 3D mouse to control a robotic arm, our approach improves precision, reduces recovery time, and enhances the resilience of automated experimental workflows.
We are working to integrate an array of tools within a widely accessible laboratory liquid handler to not only prepare samples, but also collect data, take measurements, and perform characterizations. This helps create an all-in-one ecosystem within the liquid handler to perform complete experiment protocols and collect data without the need for human inference.
A platform was developed to automate a purification process called liquid-liquid extraction and using machine learning algorithms to evaluate the quality of extraction.
We utilize dimensionality reduction techniques to understand the degeneracy nature of EIS data. This tool can be implemented in different domains to diversify the EIS dataset which will be further used for training ML models.
LISA is a robotic lab assistant that helps scientists organize and prepare their experiments, automatically tracks inventory in real-time, and reduces repetitive tasks- allowing researchers to focus more on the science itself.
Recent advances in machine learning for materials show that even simple representations, like chemical formulas without structural details, can predict properties effectively, challenging traditional physics-based intuition. This work introduces a tomographic framework based on information theory to clarify structure-property relationships and evaluates property-augmented representations across various prediction tasks.
We developed a low-cost robotic system with tactile sensors that gives robots human-like dexterity to precisely handle both rigid and flexible objects, such as wires and tubes, in complex chemical experiments. This brings us closer to fully automated labs for accelerating material discovery without the need for expensive equipment.
Automating unknown chemical reactions is challenging for multiple reasons. One challenge is to find out when a reaction is finished if something is happening at all. We developed a toolbox for spectroscopic analysis that unlocks new insights without requiring prior in-depth knowledge about the chemistry.
The objective of this study is to use a Deep Convolutional Neural Network (CNN) to denoise computed tomography (CT) scan images in order to segment crack structures and, subsequently, gain a better understanding of 2D and 3D crack propagation in various cementitious-based materials.
We developed a workflow that combines 3D molecular structures with quantum chemistry to enhance Bayesian optimization of chemical reactions, improving performance by up to 35% over traditional ligand representations.
We are creating an automated laboratory workflow that combines self-driven exploration with hyperspectral Raman and photoluminescence imaging. This allows rapid mapping of carbon nanotubes and polymer blends, accelerating functional materials characterization with minimal human intervention.
We develop a federated learning method that utilizes multi-objective optimization to enhance the training of distributed ML models on decentralized laboratories. To support real-world use, we also built a downloadable app that enables decentralized labs to deploy and train models.
Discover the future of lab automation with LADS OPC UA. This powerful new standard uses robust industrial technology (OPC UA) to seamlessly integrate diverse lab equipment and connect your lab to enterprise systems, demonstrating effortless multi-vendor control for accelerated research and development.
This project focuses on extracting chemical compound images from Google Patents and converting them into the SMILES format. Due to the lack of an official API for Google Patents, web scraping techniques were employed to collect images convert to SMILES and add to drug-like compound database
We present a framework that combines reasoning-capable language models with robotic simulations to automatically design and validate laboratory experiments. By integrating with Argonne National Laboratory's MADSci platform, our solution enables fully closed-loop autonomous laboratories that can explore any viable experiment within equipment constraints, dramatically expanding the potential search space for scientific discovery.
We developed Robotic Autonomous Imaging Surface Evaluator (RAISE), a fully closed-loop automated system that uses robotics and machine vision to perform contact angle measurements, capturing how liquids interact with surfaces. This tool speeds up the discovery of new materials and formulations by eliminating human error and running experiments on its own.
We report on the autonomous synthesis of molecular machines with real-time control for reaction monitoring and purification within the XDL programming language. By streamlining and standardizing complex syntheses, we improve reproducibility and reliability whilst enabling broader automation in chemistry thus freeing researchers to pursue more ambitious and exploratory work.
We propose an active learning method for drug discovery, acquiring labels for some samples and training a model to predict the rest. Our models prune the most difficult examples from the target set to achieve high accuracy and reduce experimental costs.
Energy transfer photocatalysis provides a versatile platform for the efficient, sustainable and rapid construction of complex molecular scaffolds. We leverage data-driven exploration and mechanistic investigation strategies to accelerate the discovery of novel synthetically valuable transformations and to better understand chemical reactivity in this emerging catalytic mode.
Leveraging automated fabrication and testing, we discovered highly active and stable electrocatalysts for producing green hydrogen
HiWAY-2-MAT is a synthesis and high-throughput characterisation platform, and one of 17 platforms being set up across France as a part of a large scale project to include automation, machine learning, and artificial intelligence in materials design. Here I describe how the platform is set up and designed, and more broadly how it fits into the larger framework of cooperative, independent platforms.
This poster introduces El Agente and El Agente Q, a multi-agent system designed to democratize quantum chemistry workflows through natural language interaction. Built on large language models, it automates complex tasks like molecular modeling, analysis, and error recovery, making cutting-edge chemistry more accessible, efficient, and adaptable for students, researchers, and other aficionados. To learn more about El Agente or pre-register to test our alpha version, visit our website www.elagente.ca.
We’ve developed an AI-driven model that helps discover high-performance fluid mixtures by learning how different molecules interact. By combining graph neural networks and Bayesian optimization, our approach accelerates the search for optimal formulations in applications like drug delivery, thermal management, and advanced materials.
The development of a self-driving MEA electrolyzer platform dramatically accelerates the discovery and optimization of CO₂ electroreduction systems by enabling autonomous, high-throughput experimentation with real-time performance feedback.
SOKE Graph is an AI-powered tool that helps researchers quickly find and organize relevant information from scientific papers about green hydrogen and clean energy. It uses smart prompts and a structured knowledge graph to highlight key concepts, saving time and making discoveries faster and easier.
We introduce ivoryOS, an open-source orchestrator for Python-based self-driving laboratories (SDLs) that automatically generates a drag-and-drop web interface from exposed SDL operations. Its plug-and-play architecture and low-code interface make it easy to share, adapt, and scale across labs—streamlining SDL development and operation for researchers.
We used artificial intelligence to design new drug-like molecules that target ABL1, a protein involved in leukemia. Our approach helps speed up the discovery of potential treatments, especially for drug-resistant cancer mutations.
Our work evaluates state-of-the-art universal machine learning potentials on structural optimization, molecular dynamics (MD), and bulk modulus and heat capacity predictions of metal-organic frameworks, a class of highly porous materials with potential applications in carbon capture, energy storage, and catalysis.
Researchers at Monash University in Australia have built a novel robotic platform, capable of making and evaluating new thin film solar cell materials in an autonomous workflow, without human intervention. This platform is part of the Australian Centre for Advanced Photovoltaics.
Can a $4 device make automated experiments tick with guaranteed precision? And can an idea in programming languages transform this hardware capability into an accessible tool for everyone to tinker with?
We have developed an automated flow system for polymer synthesis that can be controlled remotely and continuously monitored in real time using in-line NMR, IR, and UV-Vis spectroscopy. This setup enables the generation of large datasets for machine learning, accelerating the discovery of new polymer materials.
We developed a non-empirical kinetic energy functional with significantly improved energies and densities in the framework of orbital-free density functional theory (DFT). Its accuracy and its cheap quadratic computational time complexity promises a path toward accelerated exploration of molecules/materials much larger than those conventional Kohn-Sham DFT can afford.
A novel comprehensive platform for the discovery and optimization of electrochemical processes, with applications in analytical and synthetic chemistry.
Mixtures of molecules are integral to our daily experiences: from the perfumes we smell to the remedies
This project demonstrates a fully automated platform for fabricating and characterizing reverse-osmosis (RO) membranes using the nonsolvent-induced phase separation (NIPS) method. By integrating robotic solution preparation, casting, and compression-based mechanical testing, the system enables high-throughput exploration of fabrication parameters and provides rapid feedback on membrane structure and performance.
Flexible automation allows lab systems to adapt to changing experiments. We used this approach to convert a solar cell lab into one that transforms CO₂ into fuel.
Sorting through thousands of images to identify tiny insects takes immense time. We're developing cutting-edge AI that automatically detects small arthropods in bulk photos, dramatically speeding up research. This technology helps scientists analyze massive datasets, like those from the global LIFEPLAN project, leading to a faster, better understanding of Earth's biodiversity.
In this work we discuss our process for introducing automation to the modeling, fabrication and characterization of organic thin film transistors. We expand on how this has enabled new experiments in our research, such as environmentally controlled stress testing.
PEGKi orchestrates multiple parallel-run and concurrent autonomous labs, using gamified skill assessment and meta-learning to dynamically assign optimization policies and rapidly discover a target chemical mixture.
we take. However, they are complex systems that present unique modeling challenges and have been understudied by the ML community. We introduce the first data hub for chemical mixtures that provides (1) AI-solvable tasks, (2) AI-ready datasets and (3) curated benchmarks.
We've developed a high-throughput platform combining rapid catalyst synthesis and advanced evaluation methods, including machine vision analysis, to efficiently discover high-performance catalysts for producing clean hydrogen from formic acid. Integrated with artificial intelligence, this approach significantly accelerates sustainable catalyst development.
We developed a machine learning tool that predicts how long-acting injectable drugs release over time, helping researchers create better formulations faster. The model provides mechanistic insights into release behavior and demonstrates robust predictive accuracy
We show how an old AI method for making money on slot machines can cut the supercomputing costs of designing clean energy materials.
Using a carbon nanotube synthesis research robot coupled with an artificial intelligence to design experiments, we simultaneously optimized both the overall yield of material and controlled the diameter of the resulting nanotubes. We show that yield and diameter control and yield are independent outcomes and no sacrifice in yield is necessary for diameter control.
We use a smart AI method to quickly and precisely extract key information from scientific data, helping researchers characterize and discover sustainable materials faster and more efficiently.
This study developed a real-time automated workflow to detect and correct noisy data in self-driving labs using k-nearest-neighbour imputation and statistical evaluation methods. Systematic benchmarking across training data sizes and noise scenarios reveals clear performance limits, offering practical guidelines for maintaining data integrity in high-throughput discovery.
We have developed Auto-SEMEDS, a fully automated software that enables accurate, high-throughput composition analysis of powder materials using SEM-EDS. By refining SEM-EDS quantification methods and integrating machine learning analysis, Auto-SEMEDS enhances powder characterization, making it faster, more reliable, and ideal for self-driving laboratories.
The integration of Bayesian optimization with high-throughput microbial phenotyping platform enables the efficient design of synthetic microbial cultures tailored for real-world industrial applications.
HeinSight is a computer vision system that enables real-time monitoring of physical changes and provides feedback control for autonomous experimental work.
We have developed an automated flow system for polymer synthesis that can be controlled remotely and continuously monitored in real time using in-line NMR, IR, and UV-Vis spectroscopy. This setup enables the generation of large datasets for machine learning, accelerating the discovery of new polymer materials.
We demonstrate a rapid, automated material characterization tool capable of measuring all 3 figures of merit from photovoltaics to inform device quality, without the need to fully complete the fabrication process.
We developed a powerful platform using patient-derived leukemia stem cells that mimics how cancer stem cells behave and resist treatment. This platform helps researchers discover and test new features that target the root of cancer relapse, with the goal of creating more effective treatments for leukemia and other hard-to-treat cancers originating from cancer stem cells.
Deeper understanding of chemical processes is fundamental towards streamlining production across a wide range of industries. In this work we present a method for selectively monitoring different positions within an experimental sample and provide case-studies to show its utility in many different stages of chemical development.
This study explores how machine learning can optimize 3D printing parameters for polymer materials, specifically poly(caprolactone) (PCL). By analyzing 800 samples with profilometry and stereomicroscopy, we demonstrate how predictive models can speed up the iteration process, improving print quality and accelerating material innovation in additive manufacturing. Our findings highlight an efficient and scalable approach for accelerating innovation in the field.
Our self-driving lab revolutionizes battery materials research by using automation and AI to quickly and efficiently discover advanced cathode materials. By automating the entire process—from design and synthesis to testing and analysis—we significantly speed up the development of cutting-edge lithium-ion batteries, making them more effective and reliable.
We've spent 10 years on modernizing retrosynthesis tools. Chemical.AI's reaction informatics platform now supports human chemists and robotics to conduct synthesis planning, optimize reaction conditions, predict impurities, and so on.
To accelerate the discovery of useful materials, we need to minimize the time and resource consumption associated with running large number of experiments or simulations. Our work comprehensively surveys approaches to experimental planning, providing actionable insights into maximally data-efficient approaches for constructing ML material property prediction models.
We propose a federated learning method for training models with data from non-iid sources and create a downloadable application for deploying and training such models.
We present our progress in establishing two robotic laboratories, integrated fully with Argonne National Laboratory’s Workflow Execution Interface (WEI) automation software, to support an expanding portfolio of autonomous and automated biology workflows. These laboratories facilitate a range of biological applications, from basic molecular biology to bacterial growth monitoring, and will enable the design of novel peptides, proteins, and metabolic pathways for multiple biological applications.
Utilizing a variety of PATs to optimize and automate the synthesis of model predicted ligands.
The urgent need for sustainable energy solutions has positioned green hydrogen, produced through photocatalytic water splitting using sunlight, as a critical component in addressing the global climate crisis. While this carbon-free energy carrier holds immense promise, its widespread adoption hinges on overcoming significant materials science challenges, particularly in developing photocatalysts that combine high efficiency, stability, and scalability. Recent advances have revealed spin-polarized states in ferromagnetic semiconductors as a transformative avenue for optimizing photocatalytic systems, offering new mechanisms to enhance charge carrier dynamics and reaction kinetics
We are building a self-driving lab that uses automation, miniaturization, and laser spectroscopy to generate critical data to accelerate the design of sustainable chemical processes - reducing cost, time, and reliance on manual work.
This study presents a self-driving laboratory (SDL)-powered discovery of fluorene-based organic laser molecules, integrating virtual screening with automated synthesis and characterization. A total of 52 candidates were experimentally validated, including green-yellow-orange-red-NIR emitters with promising optical properties for future optoelectronic applications.
This talk or poster presents an accelerated high-throughput workflow for CO2 electrocatalyst discovery, integrating robotic synthesis, automated characterization, and a parallelized CO2 electrolyzer platform. We focus on accelerating multi-element catalyst screening to target C3 hydrocarbon production.
Our work incorporates AI to discover and design gold nanoparticles to act as highly sensitive chemical sensors in a variety of health and environmental molecular sensing applications.
We use generative AI to build 3D crystal structures of MOFs directly from PXRD patterns and synthesis precursors—bridging the gap between lab data and material discovery.
Robotic Autonomous Imaging Surface Evaluator tailored for ocular devices (RAISE-Vision) integrates customized liquid handling and computer vision to perform repetitive automated contact angle measurement on curved hydrogel surfaces, standardizing wettability analysis for the ocular device field.
By integrating the robotic synthesis, high-throughput crystal structure characterization and machine learning, we developed a model capable of predicting the synthesizability and phases of mixed-metal oxalate solid solutions to accelerate research in materials development based of metal oxalates.
This work develops a photoluminescence-based metric that connects key stability properties of 3D mixed perovskites across different form factors: from quick-and-dirty drop cast samples, to genuine, industry-quality thin-films.
We developed a self-driving laboratory system that autonomously explores new Li-ion conductor thin films. By combining sputtering synthesis and Bayesian optimization, our system identified a new amorphous thin film with one magnitude higher ionic conductivity compared to starting compound.
The Huang group from the University of Toronto have developed a novel AI-driven, microwave-based process to rapidly synthesize perovskite nanocrystals doped with Ytterbium. These nanomaterials emit intense near-infrared light by converting a single high-energy photon into two lower-energy photons (a phenomenon called quantum cutting). This technology enables on-demand tuning of light emission and could lead to more efficient solar panels (by harvesting otherwise wasted UV light) and improved infrared imaging for medical applications.
An integrated Density Functional Theory – Bayesian Optimization strategy efficiently discoveries top-performing MOFs for sustainable carbon capture.
We created a small, automated lab-on-a-chip system combined with machine learning to quickly and efficiently find the best ways to give multiple cancer drugs together or in sequence. This approach helps reduce the amount of testing needed, supports more personalized cancer treatments, and could also be used for other diseases beyond cancer.
We present a versatile and modular Self Driving Lab (SDL) capable of running parallel syntheses, with seamless integration of synthesis, purification, in-line characterization steps, data analysis after characterization, for nano and advanced materials. The SDL successfully synthesizes diverse nanomaterials with different chemistries with excellent reproducibility, demonstrating it’s potential for accelerating materials discovery.
We will share the promising research direction using AI for sustainable 2D materials and energy harvesting materials, which can be implemented in flexible electronics.
We developed a data-driven platform combining high-throughput electrochemistry and machine learning to discover and design molecular additives for interface engineering. Applied to copper electroplating, CO₂ reduction, and H/D separation, our approach reveals structure-function relationships and enables the rational tuning of interfacial properties.
This presentation highlights our recent advancements in multimodal AI and autonomous self-driving laboratories aimed at accelerating research and development in plastics and their processing.