Gaining Autonomy from the Lab to Production

Trust the Robots: Closed-loop Automated Flow and Batch Polymerizations

Mariano Nicolas Cruz Bournazou

Institute of Biotechnology
Technical University Berlin
Germany

Abstract

In the next-generation, AI-driven era of science (“the Era of Experience,” Sutton), artificial intelligence’s ability to autonomously generate informative data, and even knowledge, is already transforming process systems in biochemical and chemical engineering. At the heart of this transformation is the concept of cognitive self-driving laboratories. These laboratories are evolving into machines of autonomous discovery, enabling AI systems to propose, test, and refine hypotheses, thereby accelerating scientific breakthroughs and uncovering complex causal relationships.

As these agents execute increasingly complex laboratory, cognitive, and decision-making tasks, engineering challenges in process design and scale-up can now be addressed. We will focus our discussion on four key components:

• Data, information, and knowledge: Existing and autonomously generated data must be transformed into interoperable and machine-actionable knowledge. AI has shown the potential for thorough information management, but standards and procedures must mature to ensure automated FAIR-by-design generation of experimental results.
• Trustworthy knowledge exchange: we need platforms that enable a transparent, secure, immutable, and traceable smart-contract–based exchange of information to promote collaboration and open science.
• Scaling up: Process dynamics, plant-wide optimization, and large-scale difficulties pose specific challenges that differ significantly from existing AI applications. Although several tools have demonstrated strong performance, many issues remain unresolved; controlling dynamic processes, for example, remains an open challenge.

Tackling societal and industrial challenges: As agents gain autonomy, traditional objective and reward functions must be reshaped. Agents should consider the environmental and societal consequences of their suggestions to ensure beneficial development.

Abstract

Machine learning (ML) and artificial intelligence (AI) applications increase exponentially in daily life and in scientific discovery. Whereas ML was a speciality in the workflow of a chemist just a decade ago, it has become widespread and penetrates almost every aspect of the scientific process, either by direct application to scientific problems, or by making use of AI agents in the organization of work or screening of literature. While ML and AI have proven to be very powerful, they are, however, at all times limited in the abilities by the quality and quantity of available data. Thus, for a true expansion of knowledge in the future by using these methodologies, we will need to find ways to provide chemical data on scale.

Robotic workflows are ideal to serve in this function. Not only are robots able to provide data continuously, they are also, at least in principle, more reliable in that they work without human bias, and with high reproducibility.  Consequently, we observe the creation of more and more sophisticated robotic laboratories, ranging from simple high-throughput experimentation to integrated closed-loop analysis and automated optimization of reactions.  We will discuss the advances that have been made in the last years with respect to automated continuous flow as well as batch reactors in relation to online analytics that allow to not only synthesize polymer materials on scale, but also to provide characterization of these materials at the same time. Challenges in the process will be highlighted and needs for future developments and integrations will be highlighted.