A multimodal robotic platform for multi-element electrocatalyst discovery.

Area of Science:

• Materials Science
• Artificial Intelligence
• Chemical Engineering

Background:

• AI for Science aims to discover custom materials via real-world experiments.
• Current materials experimentation is limited by unimodal active learning, hindering AI's potential in complex experimental interpretation.
• Advances in computational prediction and automated synthesis exist but are not fully integrated with AI's interpretive capabilities.

Purpose of the Study:

• To present CRESt, a platform integrating large multimodal models with knowledge-assisted Bayesian optimization and robotic automation.
• To accelerate materials design, synthesis, characterization, and performance optimization using AI.
• To enable AI-driven anomaly diagnosis and correction in real-world experiments.

Main Methods:

• Integration of large multimodal models (chemical compositions, text embeddings, microstructural images) with knowledge-assisted Bayesian optimization and robotic automation.
• Utilizing knowledge-embedding-based search space reduction and adaptive exploration-exploitation strategies.
• Employing cameras for monitoring and vision-language models for hypothesis generation to address experimental anomalies.

Main Results:

• CRESt facilitated exploration of over 900 catalyst chemistries and 3,500 electrochemical tests in 3 months.
• Identified a state-of-the-art octonary catalyst (Pd-Pt-Cu-Au-Ir-Ce-Nb-Cr) for electrochemical formate oxidation.
• Achieved a 9.3-fold improvement in cost-specific performance compared to existing catalysts.

Conclusions:

• CRESt significantly accelerates the discovery and optimization of advanced materials through integrated AI and robotics.
• The platform demonstrates the power of multimodal AI in tackling complex experimental challenges.
• The identified octonary catalyst represents a breakthrough in formate oxidation catalysis, showcasing practical applications of AI-driven materials discovery.