A Key Challenge In Complex Visuomotor Control Is Learning Abstract Representations That Are Effective For Specifying Goals, Planning, And Generalization. To This End, We Introduce Universal Planning Networks (upn). Upns Embed Differentiable Planning Within A Goal-directed Policy. This Planning Computation Unrolls A Forward Model In A Latent Space And Infers An Optimal Action Plan Through Gradient Descent Trajectory Optimization. The Plan-by-gradient-descent Process And Its Underlying Representations Are Learned End-to-end To Directly Optimize A Supervised Imitation Learning Objective. We Find That The Representations Learned Are Not Only Effective For Goal-directed Visual Imitation Via Gradient-based Trajectory Optimization, But Can Also Provide A Metric For Specifying Goals Using Images. The Learned Representations Can Be Leveraged To Specify Distance-based Rewards To Reach New Target States For Model-free Reinforcement Learning, Resulting In Substantially More Effective Learning When Solving New Tasks Described Via Image-based Goals. We Were Able To Achieve Successful Transfer Of Visuomotor Planning Strategies Across Robots With Significantly Different Morphologies And Actuation Capabilities.
Proceedings of the 1st Annual Conference on Robot Learning on 13-15 November 2017 Published as Volume 78 by the Proceedings of Machine Learning Research on 18 October 2017. Volume Edited by: Sergey Levine Vincent Vanhoucke Ken Goldberg Series Editors: Neil D. Lawrence Mark Reid
J. Parker, A. Khoogar, and D. Goldberg. Robotics and Automation, 1989. Proceedings., 1989 IEEE International Conference on, page 271--276. IEEE, (1989)
A. Zeng, S. Song, S. Welker, J. Lee, A. Rodriguez, and T. Funkhouser. (2018)cite arxiv:1803.09956Comment: Under review at the International Conference On Intelligent Robots and Systems (IROS) 2018. Project webpage: http://vpg.cs.princeton.edu.
A. Chaabani, M. Bellamine, and M. Gasmi. International Journal on Soft Computing, Artificial Intelligence and Applications (IJSCAI), 3 (4):
17(November 2014)