Hybrid system identification using switching density networks

Michael Burke, Yordan Hristov, Subramanian Ramamoorthy

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

Abstract

Behaviour cloning is a commonly used strategy for imitation learning and can be extremely effective in constrained domains. However, in cases where the dynamics of an environment may be state dependent and varying, behaviour cloning places a burden on model capacity and the number of demonstrations required.This paper introduces switching density networks, which rely on a categorical reparametrisation for hybrid system identification. This results in a network comprising a classification layer that is followed by a regression layer. We use switching density networks to predict the parameters of hybrid control laws, which are toggled by a switching layer to produce different controller outputs, when conditioned on an input state. This work shows how switching density networks can be used for hybrid system identification in a variety of tasks, successfully identifying the key joint angle goals that make up manipulation tasks, while simultaneously learning image-based goal classifiers and regression networks that predict joint angles from images. We also show that they can cluster the phase space of an inverted pendulum, identifying the balance, spin and pump controllers required to solve this task. Switching density networks can be difficult to train, but we introduce a cross entropy regularisation loss that stabilises training.
Original languageEnglish
Title of host publication3rd Annual Conference on Robot Learning, CoRL 2019,
EditorsLeslie Pack Kaelbling, Danica Kragic, Komei Sugiura
Place of PublicationLondon UK
PublisherProceedings of Machine Learning Research (PMLR)
Pages172-181
Number of pages10
Volume100
Publication statusPublished - 2020
Externally publishedYes
EventConference on Robot Learning 2019 - Osaka, Japan
Duration: 30 Oct 20191 Nov 2019
Conference number: 3rd
http://proceedings.mlr.press/v100/ (Proceedings)
https://web.archive.org/web/20191022151748/https://www.robot-learning.org/ (Website)

Conference

ConferenceConference on Robot Learning 2019
Abbreviated titleCoRL 2019
CountryJapan
CityOsaka
Period30/10/191/11/19
Internet address

Keywords

  • Behaviour cloning
  • switching density networks
  • hybrid systems

Cite this