A data-driven approach for learning to control computers

Peter C Humphreys, David Raposo, Tobias Pohlen, Gregory Thornton, Rachita Chhaparia, Alistair Muldal, Josh Abramson, Petko Georgiev, Adam Santoro, Timothy Lillicrap
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:9466-9482, 2022.

Abstract

It would be useful for machines to use computers as humans do so that they can aid us in everyday tasks. This is a setting in which there is also the potential to leverage large-scale expert demonstrations and human judgements of interactive behaviour, which are two ingredients that have driven much recent success in AI. Here we investigate the setting of computer control using keyboard and mouse, with goals specified via natural language. Instead of focusing on hand-designed curricula and specialized action spaces, we focus on developing a scalable method centered on reinforcement learning combined with behavioural priors informed by actual human-computer interactions. We achieve state-of-the-art and human-level mean performance across all tasks within the MiniWob++ benchmark, a challenging suite of computer control problems, and find strong evidence of cross-task transfer. These results demonstrate the usefulness of a unified human-agent interface when training machines to use computers. Altogether our results suggest a formula for achieving competency beyond MiniWob++ and towards controlling computers, in general, as a human would.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-humphreys22a, title = {A data-driven approach for learning to control computers}, author = {Humphreys, Peter C and Raposo, David and Pohlen, Tobias and Thornton, Gregory and Chhaparia, Rachita and Muldal, Alistair and Abramson, Josh and Georgiev, Petko and Santoro, Adam and Lillicrap, Timothy}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {9466--9482}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/humphreys22a/humphreys22a.pdf}, url = {https://proceedings.mlr.press/v162/humphreys22a.html}, abstract = {It would be useful for machines to use computers as humans do so that they can aid us in everyday tasks. This is a setting in which there is also the potential to leverage large-scale expert demonstrations and human judgements of interactive behaviour, which are two ingredients that have driven much recent success in AI. Here we investigate the setting of computer control using keyboard and mouse, with goals specified via natural language. Instead of focusing on hand-designed curricula and specialized action spaces, we focus on developing a scalable method centered on reinforcement learning combined with behavioural priors informed by actual human-computer interactions. We achieve state-of-the-art and human-level mean performance across all tasks within the MiniWob++ benchmark, a challenging suite of computer control problems, and find strong evidence of cross-task transfer. These results demonstrate the usefulness of a unified human-agent interface when training machines to use computers. Altogether our results suggest a formula for achieving competency beyond MiniWob++ and towards controlling computers, in general, as a human would.} }
Endnote
%0 Conference Paper %T A data-driven approach for learning to control computers %A Peter C Humphreys %A David Raposo %A Tobias Pohlen %A Gregory Thornton %A Rachita Chhaparia %A Alistair Muldal %A Josh Abramson %A Petko Georgiev %A Adam Santoro %A Timothy Lillicrap %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-humphreys22a %I PMLR %P 9466--9482 %U https://proceedings.mlr.press/v162/humphreys22a.html %V 162 %X It would be useful for machines to use computers as humans do so that they can aid us in everyday tasks. This is a setting in which there is also the potential to leverage large-scale expert demonstrations and human judgements of interactive behaviour, which are two ingredients that have driven much recent success in AI. Here we investigate the setting of computer control using keyboard and mouse, with goals specified via natural language. Instead of focusing on hand-designed curricula and specialized action spaces, we focus on developing a scalable method centered on reinforcement learning combined with behavioural priors informed by actual human-computer interactions. We achieve state-of-the-art and human-level mean performance across all tasks within the MiniWob++ benchmark, a challenging suite of computer control problems, and find strong evidence of cross-task transfer. These results demonstrate the usefulness of a unified human-agent interface when training machines to use computers. Altogether our results suggest a formula for achieving competency beyond MiniWob++ and towards controlling computers, in general, as a human would.
APA
Humphreys, P.C., Raposo, D., Pohlen, T., Thornton, G., Chhaparia, R., Muldal, A., Abramson, J., Georgiev, P., Santoro, A. & Lillicrap, T.. (2022). A data-driven approach for learning to control computers. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:9466-9482 Available from https://proceedings.mlr.press/v162/humphreys22a.html.

Related Material