Rika Antonova

Logo

Active Learning of Transferable Priors, Kernels and Latent Representations for Robotics

Google Scholar Profile

Update: I defended my thesis "Transfer-Aware Kernels, Priors and Latent Spaces from Simulation to Real Robots" at the end of 2020 and will be starting a 2-year postdoc in 2021 at Stanford University as NSF/CRA CI Fellow.

Earlier: I am a PhD student at the Robotics, Perception, and Learning Lab at KTH, working in the group headed by Danica Kragic. My current research interests are mainly in the areas of Reinforcement Learning (RL) and Bayesian Optimization (BO). One recent direction was: robust embedding of simulation-based kernels to improve data efficiency of BO on hardware (e.g. for dynamic manipulation tasks, task-oriented grasping, locomotion).

Previously, I was a Masters student at the Robotics Institute at Carnegie Mellon University. Developing Bayesian optimization approaches for learning control parameters for bipedal locomotion (with Akshara Rai). During my time at CMU my advisor was Emma Brunskill and in her group I worked on developing Reinforcement Learning algorithms for education.

A few years before I was a software engineer at Google, first in the Search Personalization group and then in the Character Recognition team (developing open-source OCR engine Tesseract).


Publications, talks and preprints:

R. Antonova, M. Maydanskiy, D. Kragic, S. Devlin, K. Hofmann. Analytic Manifold Learning: Unifying and Evaluating Representations for Continuous Control. [arXiv]

Analytic Manifold Learning Talk

M. Hwasser, D. Kragic, R. Antonova. Variational Auto-Regularized Alignment for Sim-to-Real Control. IEEE International Conference on Robotics and Automation (ICRA), 2020.

ICRA2020 Talk ICRA2020 Experiments Video
Det2Stoc ICRA2020 Talk Det2Stoc Experiments Video

R. Antonova *, A. Rai *, T. Li, D. Kragic. Bayesian Optimization in Variational Latent Spaces with Dynamic Compression. Conference on Robot Learning (CoRL), 2019. [arXiv

CoRL2019 Talk CoRL2019 Experiments Video
BO-SVAE-DC CoRL2019 Talk BO-SVAE-DC Experiments

R. Antonova *, M. Kokic *, JA Stork, D. Kragic.  Global Search with Bernoulli Alternation Kernel for Task-Oriented Grasping Informed by Simulation. Conference on Robot Learning (CoRL), PMLR 87: 641-650, 2018. [arXiv]

CoRL2018 Talk CoRL2018 Experiments Video
BO-SVAE-DC Experiments BO-SVAE-DC CoRL2019 Talk

A. Rai *, R. Antonova *, F. Meier, C. Atkeson.  Using Simulation to Improve Sample Efficiency of Bayesian Optimization for Bipedal Robots. Journal of Machine Learning Research (JMLR) Bayesian Optimization Issue, (49): 1-24, 2019.

A. Rai *, R. Antonova *, S. Song, W. Martin, H. Geyer, C. Atkeson.  Bayesian Optimization Using Domain Knowledge on the ATRIAS Biped. IEEE International Conference on Robotics and Automation (ICRA), 2018 [arXivposter]

R. Antonova *, A. Rai *, C. Atkeson.  Deep Kernels for Optimizing Locomotion Controllers . 1st Conference on Robot Learning (CoRL), PMLR 78: 47-56, 2017. [arXivposter]

R. Antonova *, A. Rai *, C. Atkeson.  Sample Efficient Optimization for Learning Controllers for Bipedal Locomotion . IEEE-RAS 16th International Conference on Humanoid Robots (Humanoids), 2016. [arXiv]

ICRA2018 Talk CoRL2017 Talk
BO-SVAE-DC Experiments BO-SVAE-DC CoRL2019 Talk

R. Antonova *, S. Cruciani *. Unlocking the Potential of Simulators: Design with RL in Mind. Presented at the 3rd Multidisciplinary Conference on Reinforcement Learning and Decision Making (RLDM), 2017. [arXivposter]

RL for Pivoting Task

R. Antonova. Bayesian Optimization for Policy Search: Case Studies in Robotics and Education . Masters thesis. Robotics Institute, Carnegie Mellon University, 2016.

R. Antonova, J. Runde, M. H. Lee, E. Brunskill. Automatically Learning to Teach to the Learning Objectives.  In Proceedings of ACM Conference on Learning @ Scale, 2016. [poster]

R. Antonova, J. Runde, C. Dann, E. Brunskill. Improving the Sample Efficiency of Bayesian Optimization Policy Search for Optimal Stopping Problems. Workshop on Data-Efficient Machine Learning at ICML, 2016.

R. Antonova. Multi-task Value of Information Planning for Sequential Multi-task Bandits.  Technical Report. Robotics Institute, Carnegie Mellon University, 2016.

Bayesian optimization for education illustration Bayesian optimization for education illustration

MS Thesis Supervision

KTH MS students looking for thesis project supervision - please see: Opportunities currently available at RPL and Opportunities in other EECS divisions.

Supervising MS students has been a very good experience (links to their thesis works below). I can't take on new students at the moment due to other commitments, but I think RPL has a very good environment for MS students interested in ML/RL and robotics, so please do take a look at the above links to the currently available thesis project topics, if you are interested.

Carlo Rapisarda. Reinforcement Learning for Dexterity Transfer Between Manipulators. KTH MS Thesis, June 2019

Martin Hwasser. Converting Deterministic Simulators to Realistic Stochastic Models via Data Alignment. KTH MS Thesis, June 2019