Chemical synthesis has long been fully dependent on intuition of talented chemists. The cycle of synthesis planning, running experiments and data analysis is the state-of-the-art chemical synthesis research process. This paradigm could be replaced by robotics and machine learning (ML). Unmanned chemical system, or cyber-physical system (CPS), is an emerging paradigm that can automated for optimizing the synthesis of interest.¹⁾ Here we would like to discuss the future possibility of reduce the chemical burden in synthesis and its optimization by technologies.

CPS platform consists of five tasks that form closed cycle by calculation, simulation, ML, automated experiment execution and so on.

• Computational screening
• Task assignment
• Experiment execution
• Detection and analysis
• Self-optimization

Conceptually, iterative optimization in an automated manner supported by ML enables rapid establishment of synthetic process of the desired product.

The history of the endeavor for automated unmanned process development dates back to 1971.²⁾ High throughput aid for synthesis and analysis execution appeared in 1990s. Improvement of computation and the advent of ML finally changed the game in the early 21^st century. Supervised learning, unsupervised learning, and reinforcement learning have been incorporated to CPSs and active learning is also playing its role for low data applications.³⁾

Chemical automation is crucial for CPS development and it has also been driven by ML. Autonomous discovery system, mostly robotic one, emerged.⁴⁾ Eve is a well-known example of automated robotic system focusing on drug discovery by bioactivity analysis-based optimization.⁵⁾ Robotic systems have been elaborated by multiple robot setup like robotic arms, pipetting station, reaction media and detection/characterization modules.

Computer vision for robotic manipulation is enhancing the capability of precise control of robot arms in a real-time scale. Sensors allowed its flexible and real-time perception and accurate handling is not so hard with the current technologies.⁶⁾ Object detection is also possible and convolutional neural networks (CNNs) are playing a capital role.⁷⁾

 Automated information extraction by text-mining-based approaches, i.e. natural language processing, reduces noises and readily applicable in a routine manner now.⁸⁾

 All in all, unmanned chemistry system is on the practically applicable stage if you take a look at each step of CPS cycle.

However, there is a deep challenge of integration of technologies. From the aforementioned aspects, ML, data science, robotics, and related sciences for each purpose of system development need to be merged in a cooperative fashion to give birth to innovative and practical CPS. In the context of drug discovery, it is desirable to incorporate biological assays to optimize the synthesis toward more biologically active, lead or candidate compound creation.

Unmanned system is conceptually beautifull but it is just right now on a developing stage. It is highly probable the emergence of a paradigm shift from intuition to data-driven unmanned system. We would like to test the compatibility with our platform with CPS.

1. https://doi.org/10.3390/molecules28052232
2. https://doi.org/10.1021/ac60297a001
3. https://doi.org/10.1109/tkde.2013.165
4. https://doi.org/10.1021/acsami.9b01226
5. https://doi.org/10.1186/1759-4499-2-1
6. https://doi.org/10.1155/2015/846487
7. Girshick R., Donahue J., Darrell T., Malik J. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; Columbus, OH, USA. 20–23 June 2014.
8. https://doi.org/10.1038/s41467-020-17266-6