imagine.bib
@article{Tekden-2021-TRO,
author = {A. Tekden and A. Erdem and E. Erdem and T. Asfour and E. Ugur},
title = {Object and Relation Centric Representations for Push Effect Prediction},
journal = {Transactions on Robotics},
note = {under review},
year = {2021},
iattachedto = {Periodic-3-PartB}
}
@article{Seker-2021-NN,
author = {M.Y. Seker and A. Alperoglu and Y. Nagai and M. Asada and E. Oztop and E. Ugur},
title = {Imitation and Mirror Systems through Deep Modality Blending Networks},
journal = {Neural Networks},
note = {submitted},
year = {2021},
iattachedto = {Periodic-3-PartB}
}
@article{Ugur-2019-ROB,
author = {E. Ugur and H. Girgin},
title = {Compliant Parametric Dynamic Movement Primitives},
journal = {Robotica},
year = 2019,
iattachedto = {Periodic-2-PartB}
}
@article{Imre-2019-AB,
author = {M. Imre and E. Oztop and Y. Nagai and E. Ugur},
title = {Affordance-Based Altruistic Robotic Architecture for Human-Robot Collaboration},
journal = {Adaptive Behavior},
year = {2019},
volume = {27},
number = {4},
pages = {223--241},
doi = {10.1177/1059712318824697}
}
@article{Seker-2019-ROBOT,
author = {M.Y. Seker and A.E. Tekden and E. Ugur},
title = {Deep Effect Trajectory Prediction in Robot Manipulation},
journal = {Robotics and Autonomous Systems},
volume = {119},
pages = {173--184},
year = 2019,
iattachedto = {Periodic-2-PartB}
}
@inproceedings{Seker-2019-RSS,
title = {Conditional Neural Movement Primitives},
author = {M.Y. Seker and M. Imre and J. Piater and E. Ugur},
booktitle = {Robotics: Science and Systems (RSS)},
address = {Freiburg, Germany.},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{Imre-2019-TORK,
title = {Object Manipulation Learning via Conditional Neural Movement Primitives},
author = {Mert Imre and M.Yunus Seker ve Emre Ugur},
booktitle = {Turkiye Robotbilim Konferansi (Turkish Robotics Conference)},
year = {2019},
note = {poster publication, best poster award}
}
@inproceedings{Girgin-2018-IROS,
title = {Associative Skill Memory Models},
author = {H. Girgin and E. Ugur},
booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
pages = {6043--6048},
address = {Madrid, Spain},
year = 2018,
iattachedto = {Periodic-2-PartB}
}
@article{Zech-2019-IJRR,
title = {{Action representations in robotics: A taxonomy and
systematic Classification}},
author = {Zech, Philipp and Renaudo, Erwan and Haller, Simon
and Zhang, Xiang and Piater, Justus},
journal = {{International Journal of Robotics Research}},
year = 2019,
publisher = {SAGE},
doi = {10.1177/0278364919835020},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{Borras-2018-IROS,
author = {J\'ulia Borr\`as and Raphael Heudorfer and Samuel Rader
and Peter Kaiser and Tamim Asfour},
title = {The KIT Swiss Knife Gripper for Disassembly Tasks: A
Multi-Functional Gripper for Bimanual Manipulation
with a Single Arm},
booktitle = {IEEE/RSJ International Conference on Intelligent
Robots and Systems (IROS)},
pages = {4590--4597},
year = 2018,
iattachedto = {D7.1}
}
@article{Borras-2018-RAL-Submitted,
author = {J\'ulia Borr\`as and Raphael Heudorfer and Samuel Rader and
Peter Kaiser and Tamim Asfour},
journal = {IEEE Robotics and Automation Letters (RA-L)},
title = {{The KIT Swiss Knife Gripper for Disassembly Tasks: A Multi-Functional
Gripper for Bimanual Manipulation with a Single Arm}},
year = 2018,
iattachedto = {Periodic-1-PartB},
iobsoletedby = {Borras-2018-IROS},
note = {Submitted}
}
@inproceedings{Girgin-2017-ICRAWS,
title = {Towards Generalizable Associative Skill Memories},
author = {Girgin, Hakan and Ugur, Emre},
booktitle = {ICRA Workshop on Learning and control for autonomous manipulation
systems: the role of dimensionality reduction},
year = {2017},
iattachedto = {D4.1}
}
@inproceedings{Girgin-2018-TORK,
title = {Parametrik Dinamik Motor Primitifleri},
author = {Hakan Girgin and Emre Ugur},
booktitle = {Turkiye Robotbilim Konferansi (Turkish Robotics Conference)},
year = {2018},
iattachedto = {Periodic-1-PartB}
}
@inproceedings{Hangl-2017-IROS,
title = {{Autonomous Skill-centric Testing using Deep Learning}},
author = {Hangl, Simon and Stabinger, Sebastian and Piater, Justus},
booktitle = {{IEEE/RSJ International Conference on
Intelligent Robots and Systems}},
year = 2017,
month = 8,
day = 13,
pages = {95--102},
publisher = {IEEE},
address = {{Piscataway, NJ}},
doi = {10.1109/IROS.2017.8202143},
url = {https://iis.uibk.ac.at/public/papers/Hangl-2017-IROS.pdf},
abstract = {Software testing is an important tool to ensure
software quality. This is a hard task in robotics due to dynamic
environments and the expensive development and time-consuming
execution of test cases. Most testing approaches use model-based
and/or simulation-based testing to overcome these problems. We
propose model-free skill-centric testing in which a robot
autonomously executes skills in the real world and compares it
to previous experiences. The skills are selected by maximising
the expected information gain on the distribution of erroneous
software functions. We use deep learning to model the sensor
data observed during previous successful skill executions and to
detect irregularities. Sensor data is connected to function call
profiles such that certain misbehaviour can be related to
specific functions. We evaluate our approach in simulation and
in experiments with a KUKA LWR 4+ robot by purposefully
introducing bugs to the software. We demonstrate that these bugs
can be detected with high accuracy and without the need for the
implementation of specific tests or task-specific
models.},
iattachedto = {Periodic-1-PartB}
}
@inproceedings{Seker-2018-TORK,
title = {Sekil Baglami Kullanarak Eylem-Etki Tahmini},
author = {M. Yunus Seker and Erhan Cagirici and Emre Ugur},
booktitle = {Turkiye Robotbilim Konferansi (Turkish Robotics Conference)},
year = {2018},
iattachedto = {D4.1}
}
@inproceedings{Tekden-2018-TORK,
title = {Kaldirma Aksiyonuyla Olusan Yorungenin Uzun Kisa Donem Hafiza Modeliyle Tahmini},
author = {Ahmet E. Tekden and Emre Ugur},
booktitle = {Turkiye Robotbilim Konferansi (Turkish Robotics Conference)},
year = {2018},
iattachedto = {D4.1}
}
@article{Taniguchi-2018-arXiv-1801-08829,
title = {{Symbol Emergence in Cognitive Developmental Systems: a Survey}},
author = {Taniguchi, Tadahiro and Ugur, Emre and Hoffmann, Matej and Jamone, Lorenzo
and Nagai, Takayuki and Rosman, Benjamin and Matsuka, Toshihiko and Iwahashi, Naoto
and Oztop, Erhan and Piater, Justus and W\"{o}rg\"{o}tter, Florentin},
year = 2018,
journal = {IEEE Transactions on Cognitive and Developmental Systems},
howpublished = {arXiv:1801.08829},
url = {https://arxiv.org/abs/1801.08829},
note = {doi: 10.1109/TCDS.2018.2867772},
iattachedto = {Periodic-1-PartB}
}
@article{Zech-2017-AB,
title = {{Computational models of affordance in robotics: a
taxonomy and systematic classification}},
author = {Zech, Philipp and Haller, Simon and Rezapour Lakani, Safoura
and Ridge, Barry and Ugur, Emre and Piater, Justus},
journal = {{Adaptive Behavior}},
year = 2017,
month = 9,
day = 18,
volume = 25,
number = 5,
pages = {235--271},
publisher = {SAGE},
address = {{London, UK}},
doi = {10.1177/1059712317726357},
url = {https://iis.uibk.ac.at/public/papers/Zech-2017-AB.pdf},
abstract = {J. J. Gibson's concept of affordance, one of the
central pillars of ecological psychology, is a truly remarkable
idea that provides a concise theory of animal perception
predicated on environmental interaction. It is thus not
surprising that this idea has also found its way into robotics
research as one of the underlying theories for action
perception. The success of the theory in this regard has meant
that existing research is both abundant and diffuse by virtue of
the pursuit of multiple different paths and techniques with the
common goal of enabling robots to learn, perceive, and act upon
affordances. Up until now, there has existed no systematic
investigation of existing work in this field. Motivated by this
circumstance, in this article, we begin by defining a taxonomy
for computational models of affordances rooted in a
comprehensive analysis of the most prominent theoretical ideas
of import in the field. Subsequently, after performing a
systematic literature review, we provide a classification of
existing research within our proposed taxonomy. Finally, by both
quantitatively and qualitatively assessing the data resulting
from the classification process, we highlight gaps in the
research terrain and outline open questions for the
investigation of affordances in robotics that we believe will
help inform future work, prioritize research goals, and
potentially advance the field toward greater robot
autonomy.},
iattachedto = {Periodic-1-PartB}
}
@inproceedings{Zhou-2019-IROS,
title = {Learning Via-Point Movement Primitives with Inter- and
Extrapolation Capabilities},
author = {You Zhou and Jianfeng Gao and Tamim Asfour},
booktitle = {IEEE/RSJ International Conference on Intelligent Robots
and Systems (IROS)},
year = 2019,
organization = {IEEE},
iattachedto = {D7.1}
}
@article{suarez2019practical,
author = {Su{\'{a}}rez-Hern{\'{a}}ndez, Alejandro and Aleny{\`{a}}, Guillem and Torras, Carme},
doi = {10.1109/lra.2019.2901905},
url = {https://zenodo.org/record/3463404#.XY4cDPexXRZ},
journal = {IEEE Robotics and Automation Letters},
mendeley-groups = {research_plan},
number = {3},
pages = {2282--2288},
title = {{Practical Resolution Methods for MDPs in Robotics Exemplified with Disassembly Planning}},
volume = {4},
year = {2019},
note = {presented at ICRA 2019},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{Suarez-2020-GenPlan,
title = {Strips action discovery},
author = {Su{\'a}rez-Hern{\'a}ndez, Alejandro and Segovia-Aguas, Javier and Torras, Carme and Aleny{\`a}, Guillem},
booktitle = {Generalization in Planning Workshop at AAAI 2020},
year = {2020},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{segovia2020generalized,
title = {Generalized planning with positive and negative examples},
author = {Segovia-Aguas, Javier and Jim{\'e}nez, Sergio and Jonsson, Anders},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
volume = {34},
number = {06},
pages = {9949--9956},
year = {2020},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Suarez-2021-AAAI,
title = {Online Action Recognition},
author = {Su{\'a}rez-Hern{\'a}ndez, Alejandro and Segovia-Aguas, Javier and Torras, Carme and Aleny{\`a}, Guillem},
booktitle = {35th AAAI Conference},
year = {2021},
pages = {to appear},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{interleaving2018suarez,
author = {Suarez-Hernandez, Alejandro and Alenya, Guillem and Torras, Carme},
booktitle = {2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
doi = {10.1109/IROS.2018.8593847},
isbn = {978-1-5386-8094-0},
mendeley-groups = {msc thesis,2018-2019_disassembly_asuarez,research_plan},
month = 10,
pages = {4061--4066},
publisher = {IEEE},
title = {{Interleaving Hierarchical Task Planning and Motion Constraint Testing for Dual-Arm Manipulation}},
url = {https://zenodo.org/record/3463479#.XY4i_fexXRZ},
year = {2018},
iattachedto = {Periodic-1-PartB}
}
@inproceedings{andriella2019natural,
author = {Andriella, Antonio and Su{\'{a}}rez-Hern{\'{a}}ndez, Alejandro and Segovia-Aguas, Javier and Torras, Carme and Aleny{\`{a}}, Guillem},
booktitle = {11th International Conference on Social Robotics},
pages = {to appear},
publisher = {Springer},
title = {{Natural teaching of robot-assisted rearranging exercises for cognitive training}},
year = {2019},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{suarez2019automatic,
author = {Su{\'{a}}rez-Hern{\'{a}}ndez, Alejandro and Andriella, Antonio and Taranovi{\'c}, Aleksandar and Segovia-Aguas, Javier and Torras, Carme and Aleny{\`{a}}, Guillem},
booktitle = {IEEE ICRA},
pages = {submitted},
title = {{Automatic Learning of Cognitive Exercises for Socially Assistive Robotics}},
year = {2019},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{suarez2019leveraging,
author = {A. {Suárez-Hernández} and T. {Gaugry} and J. {Segovia-Aguas} and A. {Bernardin} and C. {Torras} and M. {Marchal} and G. {Alenyà}},
booktitle = {2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
title = {Leveraging Multiple Environments for Learning and Decision Making: a Dismantling Use Case},
year = {2020},
volume = {},
number = {},
pages = {6902-6908},
doi = {10.1109/IROS45743.2020.9341182},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Bernardin-2019-IROS,
title = {An Interactive Physically-based Model for Active Suction Phenomenon Simulation},
author = {Bernardin, Antonin and Duriez, Christian and Marchal, Maud},
booktitle = {Proc. of International Conference on Intelligent Robots and Systems (IROS)},
year = {2019},
doi = {10.1109/iros40897.2019.8967526},
publisher = {IEEE},
organization = {IEEE/RJS}
}
@inproceedings{Lagneau2019,
title = {Active Deformation through Visual Servoing of Soft Objects},
author = {Lagneau, Romain and Krupa, Alexandre and Marchal, Maud},
booktitle = {Technical Report: paper submitted to ICRA 2020},
year = {2019}
}
@inproceedings{Lagneau-2020-ICRA,
title = {Active Deformation through Visual Servoing of Soft Objects},
doi = {10.1109/icra40945.2020.9197506},
publisher = {IEEE},
author = {Romain Lagneau and Alexandre Krupa and Maud Marchal},
booktitle = {IEEE International Conference on Robotics and Automation},
year = {2020},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Sengupta-2020-ICRA,
title = {Simultaneous Tracking and Elasticity Parameter Estimation of Deformable Objects},
doi = {10.1109/ICRA40945.2020.9196770},
author = {Sengupta, Agniva and Lagneau, Romain and Krupa, Alexandre and Marchand, Eric and Marchal, Maud},
booktitle = {IEEE International Conference on Robotics and Automation},
publisher = {IEEE},
iattachedto = {Periodic-3-PartB}
}
@article{Lagneau-2020-RAL,
title = {Automatic Shape Control of Deformable Wires Based on Model-Free Visual Servoing},
doi = {10.1109/LRA.2020.3007114},
author = {Lagneau, Romain and Krupa, Alexandre and Marchal, Maud},
journal = {IEEE Robotics and Automation Letters},
volume = 5,
issue = 4,
pages = {5252--5259},
iattachedto = {Periodic-3-PartB}
}
@techreport{KIT_GripperV2Datasheet,
title = {KIT Multifunctional Robotics Gripper (V2)},
author = {Felix Hundhausen and Cornelius Klas and Tamim Asfour},
year = {2019},
institution = {Karlsruhe Institute of Technology},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{Hollenstein-2020-ICDL-Submitted,
title = {Improving Exploration of Deep Reinforcement Learning using Planning for Policy Search},
author = {Hollenstein, Jakob J. and Renaudo, Erwan and Piater, Justus},
booktitle = {Submitted to International Conference on Development and Learning and Epigenetic Robotics},
year = {2020},
noturl = {https://openreview.net/forum?id=rJe7CkrFvS},
note = {under review}
}
@techreport{RenaudoZP2019,
title = {ADES~--~Autonomous Learning of Effects and Effects Models},
author = {Renaudo, Erwan and Zech, Philipp and Piater, Justus},
year = {2019},
institution = {University of Innsbruck}
}
@article{Lueddecke-2019-RAS,
title = {Fine-grained Action Plausibility Rating},
author = {Lueddecke, Timo and Woergoetter, Florentin},
journal = {Robotics and Autonomous Systems},
year = 2020,
month = 7,
volume = 129
}
@techreport{yildiz2019cloud2mesh,
title = {Pose Estimation of 2D Recognized Parts using
Pointcloud and Mesh Processing},
author = {Yildiz E., Woergoetter F.},
year = {2019},
institution = {Georg-August University of Goettingen}
}
@inproceedings{Yildiz-2019-SITIS,
title = {DCNN-Based Screw Detection for Automated Disassembly Processes},
author = {Yildiz, Erenus and W{\"o}rg{\"o}tter, Florentin},
booktitle = {2019 15th International Conference on Signal-Image Technology \& Internet-Based Systems (SITIS)},
pages = {187--192},
year = {2019},
organization = {IEEE},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{Yildiz-2020-ROBOVIS2,
title = {DCNN-Based Screw Classification in Automated Disassembly Processes},
author = {Yildiz, Erenus and Woergoetter, Florentin},
booktitle = {International Conference on Robotics, Computer Vision and Intelligent Systems},
year = {2020},
iattachedto = {D3.2}
}
@inproceedings{Yildiz-2020-ROBOVIS1,
title = {A Visual Intelligence Scheme for Hard Drive Disassembly in Automated Recycling Routines},
author = {Yildiz, Erenus and Tobias Brinker and Erwan Renaudo and Jakob J. Hollenstein and Simon Haller-Seeber and Justus Piater and Florentin Woergoetter},
booktitle = {International Conference on Robotics, Computer Vision and Intelligent Systems},
year = {2020},
pages = {17--27},
publisher = {SciTePress},
organization = {INSTICC},
doi = {10.5220/0010016000170027},
isbn = {978-989-758-479-4},
iattachedto = {D3.2}
}
@inbook{Yildiz-2021-Springer-Submitted,
author = {Erenus Yildiz and Erwan Renaudo and Jakob Hollenstein and Justus Piater and Florentin Wörgötter.},
title = {An Extended Visual Intelligence Scheme for Disassembly in Automated Recycling Routines},
booktitle = {Robotics, Computer Vision and Intelligent Systems},
editor = {Péter Galambos and Kurosh Madani},
year = {2021},
pages = {},
publisher = {Springer},
series = {CCIS},
note = {Submitted},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Yildiz-2020-ICPRS-Submitted,
title = {DCNN-Based Wire Detection in Automated Disassembly Processes},
author = {Yildiz, Erenus and Woergoetter Florentin},
booktitle = {Submitted to International Conference on Pattern Recognition},
year = {2020},
note = {Under review},
iattachedto = {D3.2}
}
@techreport{yildiz2019gapdetection,
title = {Visual Detection of Gaps in Disassembly
Routines using Stereo-Vision},
author = {Yildiz E., Woergoetter F.},
year = {2019},
institution = {Georg-August University of Goettingen}
}
@article{KartmannPaus-2018-RAL,
author = {R. Kartmann and F. Paus and M. Grotz and T. Asfour},
title = {Extraction of Physically Plausible Support Relations to Predict and Validate Manipulation Action Effects},
pages = {3991--3998},
volume = {3},
number = {4},
journal = {IEEE Robotics and Automation Letters (RA-L)},
year = {2018},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{Ferreira2019,
author = {Fabio Ferreira and Lin Shao and Tamim Asfour and Jeannette Bohg},
title = {Learning Visual Dynamics Models of Rigid Objects using Relational Inductive Biases},
booktitle = {NeurIPS 2019 Graph Representation Learning Workshop},
pages = {0--0},
year = {2019},
url = {https://sites.google.com/view/dynamicsmodels/home?authuser=0},
iattachedto = {Periodic-2-PartB}
}
@inproceedings{Akbulut-2021-ICRA,
title = {Reward Conditioned Neural Movement Primitives for Population Based Variational Policy Optimization},
author = {M.T. Akbulut and U. Bozdogan and A. Tekden and E. Ugur},
booktitle = {International Conference on Robotics and Automation (ICRA)},
year = {2021},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Tekden-2020-ICRA,
title = {Belief Regulated Dual Propagation Nets for Learning Action Effects on Articulated Multi-Part Objects},
author = {A. E. Tekden and A. Erdem and E. Erdem and M. Imre and M.Y. Seker and E. Ugur},
booktitle = {International Conference on Robotics and Automation (ICRA)},
year = {2020},
iattachedto = {D4.2}
}
@inproceedings{Akbulut-2020-CORL,
title = {ACNMP: Flexible Skill Formation through Learning from Demonstration and Reinforcement Learning via Representation Sharing},
author = {M.T. Akbulut and E. Oztop and Y. Seker and H. Xue and A. Tekden and E. Ugur},
booktitle = {Conference on Robot Learning (CoRL)},
year = {2020},
iattachedto = {D4.2}
}
@inproceedings{Ahmetoglu-2020-BrainAndMind,
title = {Locally Weighted CNMPs for Generating Flexible Action Sequences},
author = {A. Ahmetoglu and Y. Seker and E. Oztop and M. Asada and Emre Ugur},
booktitle = {Winter Workshop on Mechanism of Brain and Mind},
year = {2020},
note = {Poster presentation},
address = {Hokkaido, Japan},
iattachedto = {D4.2}
}
@inproceedings{Seker-2020-ICRAWS,
title = {Towards a Mirror Neuron System via Dual Channel Conditional Neural Movement Primitives},
author = {M. Yunus Seker and Erhan Oztop and Mete Tuluhan Akbulut and Yukie Nagai and Minoru Asada and Emre Ugur},
booktitle = {ICRA Brain-PIL Workshop,
New advances in brain-inspired perception, interaction and learning, ICRA },
year = {2020},
note = {Poster presentation},
iattachedto = {D4.2}
}
@article{Bozdogan-2020-IJISAE,
author = {Utku Bozdogan and Emre Ugur},
title = {Learning from Multiple Demonstrations With Different Modes of Operations},
journal = {International Journal of Intelligent Systems and Applications in Engineering},
year = {2020},
volume = {8},
number = {1},
pages = {37--44},
iattachedto = {D4.2}
}
@inproceedings{Klas-2021-ICRA,
author = {Cornelius Klas and Felix Hundhausen and Jianfeng Gao and Christian R. G. Dreher and Stefan Reither and You Zhou and Tamim Asfour},
title = {The KIT Gripper: A Multi-Functional Gripper for Disassembly Tasks},
booktitle = {International Conference on Robotics and Automation (ICRA)},
year = {2021},
iattachedto = {D7.2}
}
@article{Peer-2021-NN,
title = {{Auto-tuning of Deep Neural Networks by Conflicting Layer Removal}},
author = {Peer, David and Stabinger, Sebastian and Rodr\'{\i}guez-S\'{a}nchez, Antonio},
year = 2021,
journal = {submitted to Neural Networks},
url = {https://arxiv.org/abs/2103.04331},
abstract = {Designing neural network architectures is a challenging task and knowing which specific layers of a model must be adapted to improve the performance is almost a mystery. In this paper, we introduce a novel methodology to identify layers that decrease the test accuracy of trained models. Conflicting layers are detected as early as the beginning of training. In the worst-case scenario, we prove that such a layer could lead to a network that cannot be trained at all. A theoretical analysis is provided on what is the origin of those layers that result in a lower overall network performance, which is complemented by our extensive empirical evaluation. More precisely, we identified those layers that worsen the performance because they would produce what we name conflicting training bundles. We will show that around 60\% of the layers of trained residual networks can be completely removed from the architecture with no significant increase in the test-error. We will further present a novel neural-architecture-search (NAS) algorithm that identifies conflicting layers at the beginning of the training. Architectures found by our auto-tuning algorithm achieve competitive accuracy values when compared against more complex state-of-the-art architectures, while drastically reducing memory consumption and inference time for different computer vision tasks.},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
@article{Peer-2019-cb-software-impacts,
title = {{conflicting\_bundle.py - A python module to identify problematic layers in deep neural networks}},
author = {Peer, David and Stabinger, Sebastian and Rodr\'{\i}guez-S\'{a}nchez, Antonio},
journal = {{Software Impacts}},
year = 2021,
month = 2,
volume = 7,
publisher = {Elsevier},
doi = {10.1016/j.simpa.2021.100053},
url = {http://dx.doi.org/10.1016/j.simpa.2021.100053},
abstract = {Designing neural network architectures is a challenging task and knowing which specific layers of a neural network must be adapted to improve the performance is almost a mystery. In this paper, we introduce the conflicting\_bundle.py module to identify layers that decrease the accuracy of trained networks. Therefore, this software-module helps machine-learning researchers and engineers to precisely analyze and improve neural network architectures. The same software-module can also be used to automatically create improved neural network architectures.},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Peer-2021-WACV,
title = {{Conflicting Bundles: Adapting Architectures Towards the Improved Training of Deep Neural Networks
}},
author = {Peer, David and Stabinger, Sebastian and Rodr\'{\i}guez-S\'{a}nchez, Antonio},
booktitle = {{IEEE/CVF Winter Conference on Applications of Computer Vision}},
year = 2021,
month = 1,
pages = {256--256},
url = {https://openaccess.thecvf.com/content/WACV2021/papers/Peer_Conflicting_Bundles_Adapting_Architectures_Towards_the_Improved_Training_of_Deep_WACV_2021_paper.pdf},
abstract = {Designing neural network architectures is a challenging
task and knowing which specific layers of a model must be adapted
to improve the performance is almost a mystery. In this paper, we
introduce a novel theory and metric to identify layers that
decrease the test accuracy of the trained models, this
identification is done as early as at the beginning of
training. In the worst-case, such a layer could lead to a network
that can not be trained at all. More precisely, we identified
those layers that worsen the performance because they produce
conflicting training bundles as we show in our novel theoretical
analysis, complemented by our extensive empirical studies. Based
on these findings, a novel algorithm is introduced to remove
performance decreasing layers automatically. Architectures found
by this algorithm achieve a competitive accuracy when compared
against the state-of-the-art architectures. While keeping such
high accuracy, our approach drastically reduces memory consumption
and inference time for different computer vision tasks.},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Cruz-2020-SMC,
title = {{Reconfigurable Behavior Trees: Towards an Executive Framework Meeting High-Level Decision Making and Control Layer Features}},
author = {de la Cruz, Pilar and Piater, Justus and Saveriano, Matteo},
booktitle = {{
IEEE International Conference on Systems, Man, and Cybernetics}},
year = 2020,
month = 10,
note = {To appear},
url = {https://arxiv.org/abs/2007.10663},
abstract = {Behavior Trees constitute a widespread AI tool which
has been successfully spun out in robotics. Their advantages
include simplicity, modularity, and reusability of
code. However, Behavior Trees remain a high-level decision
making engine; control features cannot be easily
integrated. This paper proposes the Reconfigurable Behavior
Trees (RBTs), an extension of the traditional BTs that considers
physical constraints from the robotic environment in the
decision making process. We endow RBTs with continuous sensory
information that permits the online monitoring of the task
execution. The resulting stimulus-driven architecture is capable
of dynamically handling changes in the executive context while
keeping the execution time low. The proposed framework is
evaluated on a set of robotic experiments. The results show that
RBTs are a promising approach for robotic task representation,
monitoring, and execution.},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Saveriano-2020-ICRA,
title = {{An Energy-based Approach to Ensure the Stability of Learned Dynamical Systems}},
author = {Saveriano, Matteo},
booktitle = {{
IEEE International Conference on Robotics and Automation}},
year = 2020,
month = 06,
publisher = {IEEE},
address = {{Piscataway, NJ}},
doi = {10.1109/ICRA40945.2020.9196978},
url = {https://arxiv.org/abs/2003.11290},
abstract = {Non-linear dynamical systems represent a compact, flexible, and robust tool for reactive motion generation. The effectiveness of dynamical systems relies on their ability to accurately represent stable motions. Several approaches have been proposed to learn stable and accurate motions from demonstration. Some approaches work by separating accuracy and stability into two learning problems, which increases the number of open parameters and the overall training time. Alternative solutions exploit single-step learning but restrict the applicability to one regression technique. This paper presents a single-step approach to learn stable and accurate motions that work with any regression technique. The approach makes energy considerations on the learned dynamics to stabilize the system at run-time while introducing small deviations from the demonstrated motion. Since the initial value of the energy injected into the system affects the reproduction accuracy, it is estimated from training data using an efficient procedure. Experiments on a real robot and a comparison on a public benchmark shows the effectiveness of the proposed approach.
},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Hollenstein-2020-KBRL,
title = {{How do Offline Measures for Exploration in Reinforcement Learning behave?}},
author = {Hollenstein, Jakob and Sayantan, Auddy and Saveriano, Matteo and Renaudo, Erwan and Piater, Justus},
booktitle = {{Knowledge Based Reinforcement Learning Workshop at IJCAI-PRICAI 2020, Yokohama, Japan
}},
year = 2021,
month = 1,
url = {https://iis.uibk.ac.at/public/papers/Hollenstein-2020-KBRL.pdf},
abstract = {
Sufficient exploration is paramount for the success of a
reinforcement learning agent. Yet, exploration is rarely
assessed in an algorithm-independent way. We compare the
behavior of three data-based, of- fline exploration metrics
described in the literature on intuitive simple distributions
and highlight prob- lems to be aware of when using them. We
propose a fourth metric, uniform relative entropy, and
implement it using either a k-nearest-neighbor or a
nearest-neighbor-ratio estimator, highlighting that the
implementation choices have a profound impact on these
measures.},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Hollenstein-2020-ACVRW,
title = {{How does explicit exploration influence Deep Reinforcement Learning?}},
author = {Hollenstein, Jakob and Renaudo, Erwan and Matteo, Saveriano and Piater, Justus},
booktitle = {{
Joint Austrian Computer Vision and Robotics Workshop}},
year = 2020,
month = 8,
pages = {29--30},
publisher = {Verlag der TU Graz},
doi = {10.3217/978-3-85125-752-6},
url = {https://iis.uibk.ac.at/public/papers/Hollenstein-2020-ACVRW.pdf},
abstract = {Most Deep Reinforcement Learning (D-RL) methods
perform local search and therefore are prone to get stuck in
non-optimal solutions. To over- come this issue, we exploit
simulation models and kinodynamic planners as exploration
mechanism in a model-based reinforcement learning method. We
show that, even on a simple toy domain, D-RL meth- ods are not
immune to local optima and require ad- ditional exploration
mechanisms. In contrast, our planning-based exploration exhibits
a better state space coverage which turns into better policies
than the ones learned via standard D-RL methods.},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
@inproceedings{Hollenstein-2019-S2R,
title = {{Evaluating Planning for Policy Search}},
author = {Hollenstein, Jakob and Piater, Justus},
booktitle = {{1st Workshop on Workshop on Closing the Reality Gap in Sim2real Transfer for Robotic Manipulation}},
year = 2019,
month = 6,
url = {https://sim2real.github.io/assets/papers/hollenstein.pdf},
abstract = {Training using randomized simulations can be seen as a
model-based reinforcement learning method. By providing a model,
this training blends planning and reinforcement learning. We
propose the use of kinodynamic planning methods as part of a
domain-randomized, model-based reinforcement learning method and
to learn in an off-policy fashion from solved planning
instances. In order for this to be an improvement, using the
planning method needs to have beneficial properties over pure
reinforcement learning. On a simple toy domain, we show an
improved state space coverage over PPO, while still also finding
trajectories with good returns.},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
@article{Peer-2019-capsule-limitations,
title = {{Limitation of capsule networks}},
author = {Peer, David and Stabinger, Sebastian and Rodr\'{\i}guez-S\'{a}nchez, Antonio},
journal = {{Pattern Recognition Letters}},
year = 2021,
month = 4,
volume = 144,
pages = {68--74},
publisher = {Elsevier},
doi = {10.1016/j.patrec.2021.01.017},
url = {http://dx.doi.org/10.1016/j.patrec.2021.01.017},
abstract = {A recently proposed method in deep learning groups multiple neurons to capsules such that each capsule represents an object or part of an object. Routing algorithms route the output of capsules from lower-level layers to upper-level layers. In this paper, we prove that state-of-the-art routing procedures decrease the expressivity of capsule networks. More precisely, it is shown that EM-routing and routing-by-agreement prevent capsule networks from distinguishing inputs and their negative counterpart. Therefore, only symmetric functions can be expressed by capsule networks, and it can be concluded that they are not universal approximators. We also theoretically motivate and empirically show that this limitation affects the training of deep capsule networks negatively. Therefore, we present an incremental improvement for state-of-the-art routing algorithms that solves the aforementioned limitation and stabilizes the training of capsule networks.},
langid = {english},
langidopts = {variant=american},
iattachedto = {Periodic-3-PartB}
}
