Note: Alonso Marco Valle has transitioned from the institute (alumni).
I am a Ph.D. student at the Intelligent Control Systems group, at the Max Planck Institute for Intelligent Systems, supervised by Prof. Dr. Sebastian Trimpe, and co-supervised by Prof. Dr. Philipp Hennig. I am also a student in the University of Tübingen, affiliated to the International Max Planck Research School for Intelligent Systems (IMPRS-IS) and associated fellow at the Max Planck ETH Center for Learning Systems.
I work in robot learning, which can be viewed as the intersection between automatic control and machine learning. During my Ph.D. I have developed special interest in Bayesian optimization with practical applications in robotics. For example, I have worked in learning robot controllers using model-free techniques.
I am also exploring how to leverage the mathematical structure of widely-known control problems and incorporate it directly in the Bayesian optimization scheme to learn robot controllers with fewer robot experiments.
In addition to this, I explore the question of until what extent should we rely on robot models, always improvable, and never perfect, and in which way robot model-free Bayesian optimization techniques can be combined with robot model-based ones from the point of view of adaptive submodularity.
Finally, I also work in Bayesian optimization with unknown constraints when the constraint threshold is unknown.
I am currently in my last year of my Ph.D. Before joining the Autonomous Motion Department in January 2016, I studied at the Polytechnic University of Barcelona, where I received my Master's degree in Automatic Control and Robotics.
Internships and collaborations
- PhD internship at Facebook Artificial Intelligence Research (FAIR), Menlo Park, California, USA, Sep-Dec 2019. Collaboration with Dr. Roberto Calandra.
- Research stay at Computational and Biological Learning Lab, University of Cambridge, UK, Feb-Apr 2019. Collaboration with Prof. Dr. José Miguel Hernández-Lobato.
Talks and posters
- Research Talk "Learning Robot Controllers using Bayesian optimization", at Computational and Biological Learning Lab, University of Cambridge, UK, April 2019
- Poster "Learning Robot Controllers under Unknown Failure Penalties using Bayesian Optimization", in Workshop Automating Robot Experiments: Manipulation and Learning, in IEEE/RSJ International Conference on Intelligent Robots and Systems (iROS), Oct 2018, Madrid (Spain)
- Invited talk "Robot controller learning using data-efficient Bayesian optimization", at Bosch Center for Artificial Intelligence, Renningen (Germany), Feb 2018
- Poster "LQR Kernels for Efficient Controller Learning", at Second Max Planck ETH Workshop on Learning Control, Feb 2018, Zurich (Switzerland)
- Poster "Bayesian Optimization for Learning Robot Control", at Google Zürich, Oct 2016
- Invited talk "Automatic Controller tuning based on Gaussian process global optimization”, at Learning and Adaptive Systems Group, ETH Zürich, Switzerland, Sep 2016
- Invited talk "Automatic LQR tuning based on Gaussian process global optimization", at First Max Planck ETH Workshop on Learning Control, Max Planck Institute for Intelligent Systems, Tübingen, Germany, Nov, 2015
- Poster "Automatic Controller Design based on Bayesian Optimization", at 1st Symposium on Intelligent Systems in Science and Industry (SISSI), Tübingen, Germany, Jul 2015
Outreach
Robot demo at Max Planck Institute for Intelligent Systems (MPI-IS): “Apollo learns to balance an inverted pole”
- At MPI-IS Machine Learning Summer School 2015 and 2017
- At MPI-IS Open house 2016
- For Cyber Valley visitors 2015—present: Dip-Ing. Volker Mornhinweg (Head of Mercedes-Benz Vans), Dip-Ing. Klaus Fröhlich (CTO of BMW), Prof. Dr. Martin Stratmann (President of Max Planck Society), Continental Chassis&Safety
- For academic visitors 2015—present: Prof. V. Kumar, Prof. N. Lawrence, Prof. P. Abeel, Prof. C. Atkenson, Prof. G. S. Sukhatme, Prof. A. Schoellig, Prof. S. Haddadin, Prof. J. Yeomans, Prof. F. Allgöwer, Prof. H. Ritter, Prof. D. Kragic
Press
- Interview "Stanfordle im Schwabenland", in Die Zeit. See article here.
Controller Tuning Machine Learning Bayesian Optimization Optimal Control Adaptive Submodularity Gaussian Process
Automatic LQR Tuning Based on Gaussian Process Global Optimization
This video demonstrates the Automatic LQR Tuning algorithm for automatic learning of feedback controllers. The auto-tuning method is based on Entropy Search, a Bayesian optimization algorithm for information-efficient global optimization.
ICRA 2017 Spotlight presentation
Presentation of our paper "Virtual vs. Real: Trading Off Simulations and Physical Experiments in Reinforcement Learning with Bayesian Optimization" published in 2017 IEEE International Conference on Robotics and Automation (ICRA), May 29 - June 3, Singapore.
Virtual vs. Real: Trading Off Simulations and Physical Experiments in Reinforcement Learning with Bayesian Optimization
Video explanation of our paper "Virtual vs. Real: Trading Off Simulations and Physical Experiments in Reinforcement Learning with Bayesian Optimization" published in 2017 IEEE International Conference on Robotics and Automation (ICRA), May 29 - June 3, Singapore.
On the Design of LQR Kernels for Efficient Controller Learning - CDC presentation
Presentation of our paper "On the Design of LQR Kernels for Efficient Controller Learning", in 56th IEEE Annual Conference on Decision and Control (CDC), Melbourne, Australia, Dec 2017
Automatic Controller Tuning on a Two-legged Robot
We challenge our automatic LQR tuning framework on a high dimensional system: A hydraulic two-legged robot performing a squatting task. Automatic LQR Tuning paper: https://is.tuebingen.mpg.de/publications/marco_icra_2016