Gated Path Planning Networks

Lisa Lee, Emilio Parisotto, Devendra Singh Chaplot, Eric Xing, Ruslan Salakhutdinov
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:2947-2955, 2018.

Abstract

Value Iteration Networks (VINs) are effective differentiable path planning modules that can be used by agents to perform navigation while still maintaining end-to-end differentiability of the entire architecture. Despite their effectiveness, they suffer from several disadvantages including training instability, random seed sensitivity, and other optimization problems. In this work, we reframe VINs as recurrent-convolutional networks which demonstrates that VINs couple recurrent convolutions with an unconventional max-pooling activation. From this perspective, we argue that standard gated recurrent update equations could potentially alleviate the optimization issues plaguing VIN. The resulting architecture, which we call the Gated Path Planning Network, is shown to empirically outperform VIN on a variety of metrics such as learning speed, hyperparameter sensitivity, iteration count, and even generalization. Furthermore, we show that this performance gap is consistent across different maze transition types, maze sizes and even show success on a challenging 3D environment, where the planner is only provided with first-person RGB images.

Cite this Paper

BibTeX
@InProceedings{pmlr-v80-lee18c, title = {Gated Path Planning Networks}, author = {Lee, Lisa and Parisotto, Emilio and Chaplot, Devendra Singh and Xing, Eric and Salakhutdinov, Ruslan}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {2947--2955}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/lee18c/lee18c.pdf}, url = {https://proceedings.mlr.press/v80/lee18c.html}, abstract = {Value Iteration Networks (VINs) are effective differentiable path planning modules that can be used by agents to perform navigation while still maintaining end-to-end differentiability of the entire architecture. Despite their effectiveness, they suffer from several disadvantages including training instability, random seed sensitivity, and other optimization problems. In this work, we reframe VINs as recurrent-convolutional networks which demonstrates that VINs couple recurrent convolutions with an unconventional max-pooling activation. From this perspective, we argue that standard gated recurrent update equations could potentially alleviate the optimization issues plaguing VIN. The resulting architecture, which we call the Gated Path Planning Network, is shown to empirically outperform VIN on a variety of metrics such as learning speed, hyperparameter sensitivity, iteration count, and even generalization. Furthermore, we show that this performance gap is consistent across different maze transition types, maze sizes and even show success on a challenging 3D environment, where the planner is only provided with first-person RGB images.} }
Endnote
%0 Conference Paper %T Gated Path Planning Networks %A Lisa Lee %A Emilio Parisotto %A Devendra Singh Chaplot %A Eric Xing %A Ruslan Salakhutdinov %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-lee18c %I PMLR %P 2947--2955 %U https://proceedings.mlr.press/v80/lee18c.html %V 80 %X Value Iteration Networks (VINs) are effective differentiable path planning modules that can be used by agents to perform navigation while still maintaining end-to-end differentiability of the entire architecture. Despite their effectiveness, they suffer from several disadvantages including training instability, random seed sensitivity, and other optimization problems. In this work, we reframe VINs as recurrent-convolutional networks which demonstrates that VINs couple recurrent convolutions with an unconventional max-pooling activation. From this perspective, we argue that standard gated recurrent update equations could potentially alleviate the optimization issues plaguing VIN. The resulting architecture, which we call the Gated Path Planning Network, is shown to empirically outperform VIN on a variety of metrics such as learning speed, hyperparameter sensitivity, iteration count, and even generalization. Furthermore, we show that this performance gap is consistent across different maze transition types, maze sizes and even show success on a challenging 3D environment, where the planner is only provided with first-person RGB images.
APA
Lee, L., Parisotto, E., Chaplot, D.S., Xing, E. & Salakhutdinov, R.. (2018). Gated Path Planning Networks. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:2947-2955 Available from https://proceedings.mlr.press/v80/lee18c.html.

Related Material