Driving school for self-driving labs

Abstract

Self-driving labs (SDLs) have emerged as a strategy for accelerating materials and chemical research. While such systems autonomously select and perform physical experiments, this does not mean that the human...

Multi-fidelity Bayesian optimization of covalent organic frameworks for xenon/krypton separations

“…Because the acquisition function paired with an optimization algorithm negates the need for humans to design the experiments inside the experiment-analysis-plan feedback loop, BO can orchestrate autonomous, "self-driving" labs [15][16][17][18][19][20][21][22] that employ automated instrumentation and/or robots to conduct a sequence of experiments with the goal of resourceefficient materials discovery and optimization.…”

“…I.e., x [i] is the vector representation of the COF, ' [i] is the delity, and y [i] is the observed Xe/Kr selectivity of the simulation conducted at iteration i. In light of this simulation data D ½n , we wish to update our prior distribution in eqn (22).…”

“…Function space view of a GP. For our problem of searching a xed pool of COFs, we are only interested in the joint distribution of the random variables listed in Y in eqn (22). However, an alternative view of the GP in eqn (5) is that it species a (prior and posterior) distribution over functions F(x,') that aim to approximate the black-box input (COF x, delity ')output (simulated Xe/Kr selectivity, y (') ) relationship underlurking the simulations-the black-box function f(x,') in eqn (26).…”

Towards a modular architecture for science factories

“…Yet the role of researchers working with SDLs is not to perform experiments themselves, but to plan, monitor, and guide SDL activities—tasks that require new skills, and thus new approaches to education and training. 72 We may also wonder whether hands-on experimental skills become less important—and, if not, how those skills are to be taught if science factories or other remote SDLs reduce opportunities for hands-on access.…”

The future of self-driving laboratories: from human in the loop interactive AI to gamification

“…One of the key elements of RL is a reward function that is made available for the algorithm during the training. [138][139][140][141] As demonstrated in Snapp et al, methods like Bayesian optimization can be easily integrated into SDL environments. However, for many real-world problems the reward functions available at the end of the experimental campaign (or aer several steps) are absent; rather the experiments are motivated by the long-term objectives.…”