Title: Gershgorin Loss Stabilizes the Recurrent Neural Network Compartment of an End-to-end Robot Learning Scheme
Authors: Lechner, Mathias 
Hasani, Ramin  
Daniela Rus 
Grosu, Radu 
Keywords: dynamical systems; Robot Learning; Continuous-time recurrent neural networks; Deep Learning; Machine Learning; Artificial Intelligence; neural networks
Issue Date: 31-Aug-2020
Book Title: 2020 IEEE International Conference on Robotics and Automation (ICRA) 
Abstract: 
Traditional robotic control suits require profound task-specific knowledge for designing, building and testing control software. The rise of Deep Learning has enabled end-to-end solutions to be learned entirely from data, requiring minimal knowledge about the application area. We design a learning scheme to train end-to-end linear dynamical systems (LDS)s by gradient descent in imitation learning robotic domains. We introduce a new regularization loss component together with a learning algorithm that improves the stability of the learned autonomous system, by forcing the eigenvalues of the internal state updates of an LDS to be negative reals. We evaluate our approach on a series of real-life and simulated robotic experiments, in comparison to linear and nonlinear Recurrent Neural Network (RNN) architectures. Our results show that our stabilizing method significantly improves test performance of LDS, enabling such linear models to match the performance of contemporary nonlinear RNN architectures. A video of the obstacle avoidance performance of our method on a mobile robot, in unseen environments, compared to other methods can be viewed at https://youtu.be/mhEsCoNao5E.
URI: http://hdl.handle.net/20.500.12708/15630
http://dx.doi.org/10.34726/242
DOI: 10.34726/242
Organisation: E191-01 - Forschungsbereich Cyber-Physical Systems 
License: Urheberrechtsschutz 1.0
Publication Type: Exhibition Contribution
Inproceedings
Konferenzbeitrag
Appears in Collections:Beitrag in Konferenzband | Conference Paper

Files in this item:

File Description SizeFormat Existing users please Login
ICRA2020_paper.pdfPaper1 MBAdobe PDFEmbargoed until September 1, 2022    Request a copy
Show full item record

Google ScholarTM

Check


Items in reposiTUm are protected by copyright, with all rights reserved, unless otherwise indicated.