JOÃO SILVÉRIO

Robotics Researcher

Robot Learning and Interaction Group
Idiap Research Institute

My goal is to create new algorithms that allow robots to learn complex motor skills from humans. In my work I exploit probabilistic machine learning techniques in combination with classical robotics concepts, (optimal) control and, recently, some reinforcement learning. I usually experiment with uni- and bi-manual manipulation applications as well as physical collaboration/interaction with humans.

Check the Publications section to learn more about my work :)

News & Updates

[New] 14/7/21 - Paper Learning from demonstration using products of experts: applications to manipulation and task prioritization accepted at IJRR! Great work by Dr. Emmanuel Pignat showing that products of experts are an efficient solution for teaching robots by imitation, with a principled way to account for task priorities. Video
[New] 2/6/21 - Our paper Ergodic exploration using tensor train: applications in insertion tasks has been accepted for publication at T-RO! We propose a promising approach to solve insertion tasks with limited sensory feedback using tensor train decomposition and ergodic exploration. Congrats to Suhan Shetty for this important step towards his PhD! Video
19/3/21 - Article Motion mappings for continuous bilateral teleoperation accepted for publication at the IEEE Robotics and Automation Letters (RA-L)! A solution based on learning diffeormorphic mappings is proposed to address workspace-mapping problems in teleoperation. Video here.
6/3/21 - The work A probabilistic framework for learning geometry-based robot manipulation skills, led by Dr. Fares Abu-Dakka, is now accepted for publication at Elsevier's Robotics and Autonoumous Systems (RAS) journal!
1/3/21 - Our paper A laser-based dual-arm system for precise control of collaborative robots has been accepted for publication at ICRA'21! Video.
17/11/2020 - Last week's IEEE Spectrum Video Friday featured the experimental results from our new pre-print!
12/7/2020 - Our paper Towards Orientation Learning and Adaptation in Cartesian Space has been accepted for publication at the IEEE Transactions on Robotics (T-RO)! Excellent work with Dr. Huang and Dr. Abu-Dakka on extending our imitation learning technique KMP to the orientation domain.
5/5/2020 - Paper accepted at R:SS 2020! Great collaboration with Thibaut Kulak in the context of his PhD research. Check Publications for details.
2/1/2020 - Workshop on Interactive Robot Learning accepted at ICRA'20!
1/7/2019 - I joined the Idiap Research Institute as a postdoctoral researcher, where I will be working with the Robot Learning and Interaction group! :)
28/6/2019 - Paper "Uncertainty-aware imitation learning using kernelized movement primitives" accepted at IROS'19! This work leverages epistemic and aleatoric uncertainties to make the robot both safe and optimal when learning from demonstrations. Check Publications for details.
23/3/2019 - Co-authored paper Kernelized Movement Primitives accepted for publication at IJRR! This work addresses motor skill learning by imitation, with a particular focus on generalization and multi-dimensional inputs.
8/10/2018 - Great news! Our paper on learning task priorities from demonstrations has been accepted at the IEEE Transactions on Robotics! Check the Publications section for details.
19/9/2018 - One paper accepted at Humanoids'18!
30/6/2018 - Three papers accepted at IROS'18! Check Publications section for details.
15/1/2018 - Good news! Our two submissions have been accepted at ICRA'18!