Differentiable Spatial Planning using Transformers

Devendra Singh Chaplot, Deepak Pathak, Jitendra Malik
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:1484-1495, 2021.

Abstract

We consider the problem of spatial path planning. In contrast to the classical solutions which optimize a new plan from scratch and assume access to the full map with ground truth obstacle locations, we learn a planner from the data in a differentiable manner that allows us to leverage statistical regularities from past data. We propose Spatial Planning Transformers (SPT), which given an obstacle map learns to generate actions by planning over long-range spatial dependencies, unlike prior data-driven planners that propagate information locally via convolutional structure in an iterative manner. In the setting where the ground truth map is not known to the agent, we leverage pre-trained SPTs in an end-to-end framework that has the structure of mapper and planner built into it which allows seamless generalization to out-of-distribution maps and goals. SPTs outperform prior state-of-the-art differentiable planners across all the setups for both manipulation and navigation tasks, leading to an absolute improvement of 7-19%.

Cite this Paper

BibTeX
@InProceedings{pmlr-v139-chaplot21a, title = {Differentiable Spatial Planning using Transformers}, author = {Chaplot, Devendra Singh and Pathak, Deepak and Malik, Jitendra}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {1484--1495}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/chaplot21a/chaplot21a.pdf}, url = {https://proceedings.mlr.press/v139/chaplot21a.html}, abstract = {We consider the problem of spatial path planning. In contrast to the classical solutions which optimize a new plan from scratch and assume access to the full map with ground truth obstacle locations, we learn a planner from the data in a differentiable manner that allows us to leverage statistical regularities from past data. We propose Spatial Planning Transformers (SPT), which given an obstacle map learns to generate actions by planning over long-range spatial dependencies, unlike prior data-driven planners that propagate information locally via convolutional structure in an iterative manner. In the setting where the ground truth map is not known to the agent, we leverage pre-trained SPTs in an end-to-end framework that has the structure of mapper and planner built into it which allows seamless generalization to out-of-distribution maps and goals. SPTs outperform prior state-of-the-art differentiable planners across all the setups for both manipulation and navigation tasks, leading to an absolute improvement of 7-19%.} }
Endnote
%0 Conference Paper %T Differentiable Spatial Planning using Transformers %A Devendra Singh Chaplot %A Deepak Pathak %A Jitendra Malik %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-chaplot21a %I PMLR %P 1484--1495 %U https://proceedings.mlr.press/v139/chaplot21a.html %V 139 %X We consider the problem of spatial path planning. In contrast to the classical solutions which optimize a new plan from scratch and assume access to the full map with ground truth obstacle locations, we learn a planner from the data in a differentiable manner that allows us to leverage statistical regularities from past data. We propose Spatial Planning Transformers (SPT), which given an obstacle map learns to generate actions by planning over long-range spatial dependencies, unlike prior data-driven planners that propagate information locally via convolutional structure in an iterative manner. In the setting where the ground truth map is not known to the agent, we leverage pre-trained SPTs in an end-to-end framework that has the structure of mapper and planner built into it which allows seamless generalization to out-of-distribution maps and goals. SPTs outperform prior state-of-the-art differentiable planners across all the setups for both manipulation and navigation tasks, leading to an absolute improvement of 7-19%.
APA
Chaplot, D.S., Pathak, D. & Malik, J.. (2021). Differentiable Spatial Planning using Transformers. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:1484-1495 Available from https://proceedings.mlr.press/v139/chaplot21a.html.

Related Material