Value propagation networks

Conference item

We present Value Propagation (VProp), a parameter-efficient differentiable planning module built on Value Iteration which can successfully be trained in a reinforcement learning fashion to solve unseen tasks, has the capability to generalize to larger map sizes, and can learn to navigate in dynamic environments. We evaluate on configurations of MazeBase grid-worlds, with randomly generated environments of several different sizes. Furthermore, we show that the module enables to learn to plan when the environment also includes stochastic elements, providing a cost-efficient learning system to build low-level size-invariant planners for a variety of interactive navigation problems.

Authors: Nardelli, N; Synnaeve, G; Lin, Z; Kohli, P; Torr, PHS

• Publication status: Published
• Peer review status: Reviewed (other)
• Publisher: OpenReview
• Host title: Proceedings of the International Conference on Learning Representations (ICLR 2019)
• Article number: 1308
• Publication date: 2019-03-29
• Acceptance date: 2019-01-01
• Event title: 7th International Conference on Learning Representations (ICLR 2019)
• Event location: New Orleans, Louisiana, USA
• Event website: https://iclr.cc/Conferences/2019
• Event start date: 2019-05-06
• Event end date: 2019-05-09
• Language: English
• Keywords: learning to plan; reinforcement learning; convolutional neural networks; navigation; value iteration