ChemOS: An orchestration software to democratize autonomous discovery

Abstract

The current Edisonian approach to discovery requires up to two decades of fundamental and applied research for materials technologies to reach the market. Such a slow and capital-intensive turnaround calls for disruptive strategies to expedite innovation. Self-driving laboratories have the potential to provide the means to revolutionize experimentation by empowering automation with artificial intelligence to enable autonomous discovery. However, the lack of adequate software solutions significantly impedes the development of self-driving laboratories. In this paper, we make progress towards addressing this challenge, and we propose and develop an implementation of ChemOS; a portable, modular and versatile software package which supplies the structured layers necessary for the deployment and operation of self-driving laboratories. ChemOS facilitates the integration of automated equipment, and it enables remote control of automated laboratories. ChemOS can operate at various degrees of autonomy; from fully unsupervised experimentation to actively including inputs and feedbacks from researchers into the experimentation loop. The flexibility of ChemOS provides a broad range of functionality as demonstrated on five applications, which were executed on different automated equipment, highlighting various aspects of the software package.

Fig 1. Representation of the modules composing ChemOS.

This scheme highlights the modularity and the independence of the six modules, which are (i) global learning procedures, (ii) automated robotic platforms, (iii) characterization equipment, (iv) databases handling and management, (v) intuitive interfaces for researchers, and (vi) online analysis. The central workflow manager, ChemOS, is depicted in yellow. The required modules to reach autonomy in the discovery process are presented in dark orange.

Fig 2

(a) The three dyes used in this experiment: E102 (yellow), E129 (red), and E133 (blue). (b) Picture of the solutions obtained with the 12 exploitation points from Phoenics (c) Picture of the in-house built robot. (d) Maximum norm distance (loss) between the achieved normalized RGB color code and the target RGB color code for the 25 experiments. Each panel corresponds to the learning procedure in-use: Phoenics, SMAC, or spearmint. For the Phoenics algorithm, red denotes a bias towards exploration, and blue a bias towards exploitation. Note that in exploration mode, Phoenics samples parameters to gather knowledge where the algorithms has only limited information. In exploitation mode, however, Phoenics makes the best decision given current information and suggests parameters in the vicinity of the best performing experiment.

Fig 3

Results from the pH (A) and density (B) experiments. Both the loss (I) and the contribution of the starting materials to the produced mixture (II) are reported. (C) Example of a dialogue between ChemOS and researchers.

Fig 4. Representation of the ChemOS pipeline while screening the Tequila Sunrise space.

Fig 5

(A) Schematic of the flow path for the sampling sequence used with the N9 robotic platform. The six parameters (P1-P6) are color coded to illustrate the effect they have on the sampling sequence. The yellow shade highlights the arm valve, and the grey shade the HPLC valve. (B) Example of logging messages from ChemOS. (C) Side and (D) top view of the robotic hardware. Lower panels: Results from the autonomous calibration of an HPLC setup maximizing the magnitude of the response. ChemOS performed autocalibration with four different learning procedures (Phoenics, SMAC, Spearmint and Uniform). Upper panels display the achieved peak areas, i.e. magnitudes of response. Lower panels display the distributions of pair-wise distances between sampled parameter points computed with the L2 norm. For the Phoenics algorithm, red denotes exploration, and blue exploitation.

Fig 6. Average performance of the experiment planning strategies leveraged by ChemOS.