Instruction-driven history-aware policies for robotic manipulations

Pierre-Louis Guhur, Shizhe Chen, Ricardo Garcia Pinel, Makarand Tapaswi, Ivan Laptev, Cordelia Schmid
Proceedings of The 6th Conference on Robot Learning, PMLR 205:175-187, 2023.

Abstract

In human environments, robots are expected to accomplish a variety of manipulation tasks given simple natural language instructions. Yet, robotic manipulation is extremely challenging as it requires fine-grained motor control, long-term memory as well as generalization to previously unseen tasks and environments. To address these challenges, we propose a unified transformer-based approach that takes into account multiple inputs. In particular, our transformer architecture integrates (i) natural language instructions and (ii) multi-view scene observations while (iii) keeping track of the full history of observations and actions. Such an approach enables learning dependencies between history and instructions and improves manipulation precision using multiple views. We evaluate our method on the challenging RLBench benchmark and on a real-world robot. Notably, our approach scales to 74 diverse RLBench tasks and outperforms the state of the art. We also address instruction-conditioned tasks and demonstrate excellent generalization to previously unseen variations.

Cite this Paper


BibTeX
@InProceedings{pmlr-v205-guhur23a, title = {Instruction-driven history-aware policies for robotic manipulations}, author = {Guhur, Pierre-Louis and Chen, Shizhe and Pinel, Ricardo Garcia and Tapaswi, Makarand and Laptev, Ivan and Schmid, Cordelia}, booktitle = {Proceedings of The 6th Conference on Robot Learning}, pages = {175--187}, year = {2023}, editor = {Liu, Karen and Kulic, Dana and Ichnowski, Jeff}, volume = {205}, series = {Proceedings of Machine Learning Research}, month = {14--18 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v205/guhur23a/guhur23a.pdf}, url = {https://proceedings.mlr.press/v205/guhur23a.html}, abstract = {In human environments, robots are expected to accomplish a variety of manipulation tasks given simple natural language instructions. Yet, robotic manipulation is extremely challenging as it requires fine-grained motor control, long-term memory as well as generalization to previously unseen tasks and environments. To address these challenges, we propose a unified transformer-based approach that takes into account multiple inputs. In particular, our transformer architecture integrates (i) natural language instructions and (ii) multi-view scene observations while (iii) keeping track of the full history of observations and actions. Such an approach enables learning dependencies between history and instructions and improves manipulation precision using multiple views. We evaluate our method on the challenging RLBench benchmark and on a real-world robot. Notably, our approach scales to 74 diverse RLBench tasks and outperforms the state of the art. We also address instruction-conditioned tasks and demonstrate excellent generalization to previously unseen variations. } }
Endnote
%0 Conference Paper %T Instruction-driven history-aware policies for robotic manipulations %A Pierre-Louis Guhur %A Shizhe Chen %A Ricardo Garcia Pinel %A Makarand Tapaswi %A Ivan Laptev %A Cordelia Schmid %B Proceedings of The 6th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2023 %E Karen Liu %E Dana Kulic %E Jeff Ichnowski %F pmlr-v205-guhur23a %I PMLR %P 175--187 %U https://proceedings.mlr.press/v205/guhur23a.html %V 205 %X In human environments, robots are expected to accomplish a variety of manipulation tasks given simple natural language instructions. Yet, robotic manipulation is extremely challenging as it requires fine-grained motor control, long-term memory as well as generalization to previously unseen tasks and environments. To address these challenges, we propose a unified transformer-based approach that takes into account multiple inputs. In particular, our transformer architecture integrates (i) natural language instructions and (ii) multi-view scene observations while (iii) keeping track of the full history of observations and actions. Such an approach enables learning dependencies between history and instructions and improves manipulation precision using multiple views. We evaluate our method on the challenging RLBench benchmark and on a real-world robot. Notably, our approach scales to 74 diverse RLBench tasks and outperforms the state of the art. We also address instruction-conditioned tasks and demonstrate excellent generalization to previously unseen variations.
APA
Guhur, P., Chen, S., Pinel, R.G., Tapaswi, M., Laptev, I. & Schmid, C.. (2023). Instruction-driven history-aware policies for robotic manipulations. Proceedings of The 6th Conference on Robot Learning, in Proceedings of Machine Learning Research 205:175-187 Available from https://proceedings.mlr.press/v205/guhur23a.html.

Related Material