Project members: Kefeng Huang, Jonathon Pipe, Dr.ir. Jihong Zhu, Prof. Andy Tyrrell and Prof. Ian Fairlamb.

Chemistry experimentation is fundamental to scientific research and development. Nevertheless, it often entails time-consuming, labour-intensive processes that are susceptible to human error. Robotic systems hold immense promise in transforming this field by automating and streamlining chemistry laboratory experiments. However, due to the intricate dexterity required, programming the behaviour through explicit instructions can be a formidable challenge.

This research endeavour seeks to tap into human expertise and explore the potential of utilising Learning from Demonstrations (LfD) techniques. This approach empowers robots to learn directly from human demonstrations, enabling them to perform chemistry experiments with increased efficiency and precision. The study will specifically investigate the application of Schlenk line techniques, transferring knowledge and extensive hands-on experience from humans to robot through kinesthetic teaching or teleoperation.

List of Research

1. Chemistry Manipulation Taxonomy

Chemistry Manipulation Taxonomy

paper currently under revision for nature communication chemistry

Laboratory automation has made well-defined experimental protocols increasingly executable by machines, yet the physical manipulations surrounding those protocols remain difficult to generalize. Setup, transfer, adjustment, assembly, and cleanup operations form a long tail of context-dependent actions that are typically handled manually or implemented as bespoke robotic skills for individual workflows and platforms.

TARMAC is an empirical taxonomy of laboratory actions grounded in the analysis of instructional chemistry practice. Rather than prescribing a control framework, it emerges from a bottom-up decomposition of real laboratory manipulations into physically meaningful primitives organized by wrench dependence, actuation directness, and motion periodicity.

Across the experimental contexts examined, the majority of actions can be expressed as compositions of a finite and reusable set of primitives. This structure provides a basis for organizing, comparing, and reusing manipulation capabilities across experimental workflows.

More information: https://tarmac-paper.github.io/#resources