Posters can be found at each of our venues throughout the conference.
Bahen Centre for Information Technology
Myhal Centre for Engineering Innovation and Entrepreneurship
Posters can be found at each of our venues throughout the conference.
Solution processable molecular semiconductors are an attractive option for low-cost, and tuneable large area optoelectronic devices such as solar cells, photodetectors, and lighting panels. To date, the choice of materials has been driven by a trial-and-error approach rather than by computational-led design. In this work, we developed an AI-based material discovery approach to efficiently explore the chemical space of large oligomers and find potential component that can realise efficient photoelectric conversion with targeted optical properties.
Liquid handling is crucial in chemistry laboratories, but typical robot grippers cannot manipulate standard pipettes. We propose a 3D-printed digital pipette for robots to perform accurate liquid handling with two-finger robot grippers. Experimental results demonstrate that robots equipped with the digital pipette could achieve accurate liquid transfer similar to commercially available devices.
Inexpensive machine learning potentials are increasingly being used to speed up simulations of materials by iteratively predicting and applying interatomic forces. In these settings, it is crucial to detect when predictions are unreliable to avoid wrong or misleading results. We present a complete framework for training and recalibrating uncertainty aware models to produce accurate predictions of energy and forces with calibrated uncertainty estimates.
Discover how robotics and automation are revolutionizing material science research. Our team has developed a simulation environment that replicates the movements within a material acceleration platform. By integrating AI and robotics, we explore how to accelerate the synthesis of new materials through physics-based robot simulations. With our self-driving laboratory and autonomous mobile robots, we are pushing the boundaries of material science, optimizing configurations, and paving the way for sustainable solutions in renewable energy.
We introduce a new method for generating compact yet accurate feature vectors of chemical systems for use with machine learning models. Our method minimizes the feature vector dimensionality for use with quantum machine learning models and achieve unprecedented chemical compound space sampling rate which can be used for large scale screening of materials and drugs.
Our automated modular probe station aimed at efficiently characterizing the electrical, optoelectrical and thermoelectrical properties of thin films on large area.
Our research tackles two challenges in machine learning applied to materials science: robustness and information redundancy. We show that big data doesn't guarantee accurate predictions and demonstrate tools to improve the robustness of machine learning prediction. We reveal that a large portion of data can be removed without compromising model performance. Our findings provide valuable insights for more efficient and reliable materials databases and machine learning models.
Self-driving labs (SDLs) can collect a vast number of images per experiment. However, a human is often needed to analyze these images, which creates a bottleneck. This poster will show how the Berlinguette Group is using computer vision and machine learning to reduce the need for human intervention in SDL workflows, which translates to increased SDL efficiency.
In this study, we present a machine learning (ML)-based method for exploring vast material spaces in search of compounds with specific properties. We demonstrate that using ML-generated projected density of states fingerprints allows us to find materials with similar properties despite their significant structural or compositional differences. The efficiency of the ML model enables scaling this approach to material spaces containing over 100,000 materials, with a low computational cost.
Discover how the Berlinguette Group uses flexible automation to build self-driving labs for accelerated materials discovery. Our approach embraces modularity, which means we can readily adapt to new research directions by reconfiguring our platforms for different experiment workflows. This versatility can enable efficient exploration of many types of materials.
Accelerated materials discovery campaigns require selecting from thousands of possible precursors and chemicals to find ones that not only achieve desired material properties, but also meet safety, economic, and engineering constraints to enable breakthrough applications. Applying a pipeline of established cheminformatics and engineering criteria to outline and prioritize a chemical search space is an effective approach to focus automated high throughput experimentation campaigns on compounds that are likely to meet both property and engineering requirements.
Reducing the Training data needed for modern machine learning methods, through the use of hierarchical methods.
Generating low-level robot task plans from high-level natural language instructions is a challenge. Existing large language models lack verified accuracy and struggle with limited domain-specific language data. CLAIRify is a novel approach that combines iterative prompting and program verification to ensure syntactically valid plans. It provides effective guidance and achieves state-of-the-art results in planning chemistry experiments, demonstrating successful execution on a real robot.
This poster will showcase a platform for accelerated fabrication and testing of materials used in electrochemical reactors that convert CO2 into valuable products.
We present a framework that uses computer vision to address specific pain points in the characterization of perovskite semiconductors. With this approach, we automate and accelerate the measurement and computation of both chemical and optoelectronic properties of perovskites.
We present our recent progress in converting our self-driving laboratory, primarily used for biological applications, into an open-source, scalable, and AI-integrated robotic platform using Argonne National Laboratory’s Workflow Execution Interface (WEI) integration software. By integrating all our robotic equipment with WEI, instead of an expensive industry solution, we have increased our platform’s flexibility and positioned ourselves to participate in the ongoing effort of standardizing robotic integration across Argonne and the broader self-driving lab scientific community.
Revolutionize your screening capabilities with AFION lab. Our innovative tech combines microfluidic synthesis, real-time spectroscopy, and state-of-the-art machine learning. Explore vast chemical space for nanoparticle synthesis, rapidly identifying optimal reaction conditions. From drug delivery to renewable energy, AFION streamlines screening, saving time and resources. Unleash AFION lab's power to propel your research with efficient screening for diverse synthesis objectives.
To accelerate the discovery of new materials, a robot framework for autonomous chemistry experiments is proposed. The framework autonomously plans multi-step actions and motions from a high-level description of the experiment and perception of the lab workspace. The robot interacts with lab equipment, incorporating constrained task and motion planning to prevent collisions and spillage. The framework successfully performs two fundamental chemistry experiments: solubility measurement and recrystallization.
Solution processable molecular semiconductors are an attractive option for low-cost, and tuneable large area optoelectronic devices such as solar cells, photodetectors, and lighting panels. To date, the choice of materials has been driven by a trial-and-error approach rather than by computational-led design. In this work, we developed an AI-based material discovery approach to efficiently explore the chemical space of large oligomers and find potential component that can realise efficient photoelectric conversion with targeted optical properties.
We introduce a new method for generating compact yet accurate feature vectors of chemical systems for use with machine learning models. Our method minimizes the feature vector dimensionality for use with quantum machine learning models and achieve unprecedented chemical compound space sampling rate which can be used for large scale screening of materials and drugs.
In this study, we present a machine learning (ML)-based method for exploring vast material spaces in search of compounds with specific properties. We demonstrate that using ML-generated projected density of states fingerprints allows us to find materials with similar properties despite their significant structural or compositional differences. The efficiency of the ML model enables scaling this approach to material spaces containing over 100,000 materials, with a low computational cost.
Reducing the Training data needed for modern machine learning methods, through the use of hierarchical methods.
This poster will showcase a platform for accelerated fabrication and testing of materials used in electrochemical reactors that convert CO2 into valuable products.
Revolutionize your screening capabilities with AFION lab. Our innovative tech combines microfluidic synthesis, real-time spectroscopy, and state-of-the-art machine learning. Explore vast chemical space for nanoparticle synthesis, rapidly identifying optimal reaction conditions. From drug delivery to renewable energy, AFION streamlines screening, saving time and resources. Unleash AFION lab's power to propel your research with efficient screening for diverse synthesis objectives.
Liquid handling is crucial in chemistry laboratories, but typical robot grippers cannot manipulate standard pipettes. We propose a 3D-printed digital pipette for robots to perform accurate liquid handling with two-finger robot grippers. Experimental results demonstrate that robots equipped with the digital pipette could achieve accurate liquid transfer similar to commercially available devices.
Inexpensive machine learning potentials are increasingly being used to speed up simulations of materials by iteratively predicting and applying interatomic forces. In these settings, it is crucial to detect when predictions are unreliable to avoid wrong or misleading results. We present a complete framework for training and recalibrating uncertainty aware models to produce accurate predictions of energy and forces with calibrated uncertainty estimates.
Discover how robotics and automation are revolutionizing material science research. Our team has developed a simulation environment that replicates the movements within a material acceleration platform. By integrating AI and robotics, we explore how to accelerate the synthesis of new materials through physics-based robot simulations. With our self-driving laboratory and autonomous mobile robots, we are pushing the boundaries of material science, optimizing configurations, and paving the way for sustainable solutions in renewable energy.
Our automated modular probe station aimed at efficiently characterizing the electrical, optoelectrical and thermoelectrical properties of thin films on large area.
Self-driving labs (SDLs) can collect a vast number of images per experiment. However, a human is often needed to analyze these images, which creates a bottleneck. This poster will show how the Berlinguette Group is using computer vision and machine learning to reduce the need for human intervention in SDL workflows, which translates to increased SDL efficiency.
Discover how the Berlinguette Group uses flexible automation to build self-driving labs for accelerated materials discovery. Our approach embraces modularity, which means we can readily adapt to new research directions by reconfiguring our platforms for different experiment workflows. This versatility can enable efficient exploration of many types of materials.
Accelerated materials discovery campaigns require selecting from thousands of possible precursors and chemicals to find ones that not only achieve desired material properties, but also meet safety, economic, and engineering constraints to enable breakthrough applications. Applying a pipeline of established cheminformatics and engineering criteria to outline and prioritize a chemical search space is an effective approach to focus automated high throughput experimentation campaigns on compounds that are likely to meet both property and engineering requirements.
Generating low-level robot task plans from high-level natural language instructions is a challenge. Existing large language models lack verified accuracy and struggle with limited domain-specific language data. CLAIRify is a novel approach that combines iterative prompting and program verification to ensure syntactically valid plans. It provides effective guidance and achieves state-of-the-art results in planning chemistry experiments, demonstrating successful execution on a real robot.
We present a framework that uses computer vision to address specific pain points in the characterization of perovskite semiconductors. With this approach, we automate and accelerate the measurement and computation of both chemical and optoelectronic properties of perovskites.
We present our recent progress in converting our self-driving laboratory, primarily used for biological applications, into an open-source, scalable, and AI-integrated robotic platform using Argonne National Laboratory’s Workflow Execution Interface (WEI) integration software. By integrating all our robotic equipment with WEI, instead of an expensive industry solution, we have increased our platform’s flexibility and positioned ourselves to participate in the ongoing effort of standardizing robotic integration across Argonne and the broader self-driving lab scientific community.
To accelerate the discovery of new materials, a robot framework for autonomous chemistry experiments is proposed. The framework autonomously plans multi-step actions and motions from a high-level description of the experiment and perception of the lab workspace. The robot interacts with lab equipment, incorporating constrained task and motion planning to prevent collisions and spillage. The framework successfully performs two fundamental chemistry experiments: solubility measurement and recrystallization.
Our research tackles two challenges in machine learning applied to materials science: robustness and information redundancy. We show that big data doesn't guarantee accurate predictions and demonstrate tools to improve the robustness of machine learning prediction. We reveal that a large portion of data can be removed without compromising model performance. Our findings provide valuable insights for more efficient and reliable materials databases and machine learning models.
Solution processable molecular semiconductors are an attractive option for low-cost, and tuneable large area optoelectronic devices such as solar cells, photodetectors, and lighting panels. To date, the choice of materials has been driven by a trial-and-error approach rather than by computational-led design. In this work, we developed an AI-based material discovery approach to efficiently explore the chemical space of large oligomers and find potential component that can realise efficient photoelectric conversion with targeted optical properties.
Liquid handling is crucial in chemistry laboratories, but typical robot grippers cannot manipulate standard pipettes. We propose a 3D-printed digital pipette for robots to perform accurate liquid handling with two-finger robot grippers. Experimental results demonstrate that robots equipped with the digital pipette could achieve accurate liquid transfer similar to commercially available devices.
Inexpensive machine learning potentials are increasingly being used to speed up simulations of materials by iteratively predicting and applying interatomic forces. In these settings, it is crucial to detect when predictions are unreliable to avoid wrong or misleading results. We present a complete framework for training and recalibrating uncertainty aware models to produce accurate predictions of energy and forces with calibrated uncertainty estimates.
Discover how robotics and automation are revolutionizing material science research. Our team has developed a simulation environment that replicates the movements within a material acceleration platform. By integrating AI and robotics, we explore how to accelerate the synthesis of new materials through physics-based robot simulations. With our self-driving laboratory and autonomous mobile robots, we are pushing the boundaries of material science, optimizing configurations, and paving the way for sustainable solutions in renewable energy.
We introduce a new method for generating compact yet accurate feature vectors of chemical systems for use with machine learning models. Our method minimizes the feature vector dimensionality for use with quantum machine learning models and achieve unprecedented chemical compound space sampling rate which can be used for large scale screening of materials and drugs.
Our automated modular probe station aimed at efficiently characterizing the electrical, optoelectrical and thermoelectrical properties of thin films on large area.
Our research tackles two challenges in machine learning applied to materials science: robustness and information redundancy. We show that big data doesn't guarantee accurate predictions and demonstrate tools to improve the robustness of machine learning prediction. We reveal that a large portion of data can be removed without compromising model performance. Our findings provide valuable insights for more efficient and reliable materials databases and machine learning models.
Self-driving labs (SDLs) can collect a vast number of images per experiment. However, a human is often needed to analyze these images, which creates a bottleneck. This poster will show how the Berlinguette Group is using computer vision and machine learning to reduce the need for human intervention in SDL workflows, which translates to increased SDL efficiency.
In this study, we present a machine learning (ML)-based method for exploring vast material spaces in search of compounds with specific properties. We demonstrate that using ML-generated projected density of states fingerprints allows us to find materials with similar properties despite their significant structural or compositional differences. The efficiency of the ML model enables scaling this approach to material spaces containing over 100,000 materials, with a low computational cost.
Discover how the Berlinguette Group uses flexible automation to build self-driving labs for accelerated materials discovery. Our approach embraces modularity, which means we can readily adapt to new research directions by reconfiguring our platforms for different experiment workflows. This versatility can enable efficient exploration of many types of materials.
Accelerated materials discovery campaigns require selecting from thousands of possible precursors and chemicals to find ones that not only achieve desired material properties, but also meet safety, economic, and engineering constraints to enable breakthrough applications. Applying a pipeline of established cheminformatics and engineering criteria to outline and prioritize a chemical search space is an effective approach to focus automated high throughput experimentation campaigns on compounds that are likely to meet both property and engineering requirements.
Reducing the Training data needed for modern machine learning methods, through the use of hierarchical methods.
Generating low-level robot task plans from high-level natural language instructions is a challenge. Existing large language models lack verified accuracy and struggle with limited domain-specific language data. CLAIRify is a novel approach that combines iterative prompting and program verification to ensure syntactically valid plans. It provides effective guidance and achieves state-of-the-art results in planning chemistry experiments, demonstrating successful execution on a real robot.
This poster will showcase a platform for accelerated fabrication and testing of materials used in electrochemical reactors that convert CO2 into valuable products.
We present a framework that uses computer vision to address specific pain points in the characterization of perovskite semiconductors. With this approach, we automate and accelerate the measurement and computation of both chemical and optoelectronic properties of perovskites.
We present our recent progress in converting our self-driving laboratory, primarily used for biological applications, into an open-source, scalable, and AI-integrated robotic platform using Argonne National Laboratory’s Workflow Execution Interface (WEI) integration software. By integrating all our robotic equipment with WEI, instead of an expensive industry solution, we have increased our platform’s flexibility and positioned ourselves to participate in the ongoing effort of standardizing robotic integration across Argonne and the broader self-driving lab scientific community.
Revolutionize your screening capabilities with AFION lab. Our innovative tech combines microfluidic synthesis, real-time spectroscopy, and state-of-the-art machine learning. Explore vast chemical space for nanoparticle synthesis, rapidly identifying optimal reaction conditions. From drug delivery to renewable energy, AFION streamlines screening, saving time and resources. Unleash AFION lab's power to propel your research with efficient screening for diverse synthesis objectives.
To accelerate the discovery of new materials, a robot framework for autonomous chemistry experiments is proposed. The framework autonomously plans multi-step actions and motions from a high-level description of the experiment and perception of the lab workspace. The robot interacts with lab equipment, incorporating constrained task and motion planning to prevent collisions and spillage. The framework successfully performs two fundamental chemistry experiments: solubility measurement and recrystallization.