Generative Models for Learning Robot Manipulation Skills from Humans
Event details
| Date | 12.12.2017 |
| Hour | 10:00 › 12:00 |
| Speaker | Ajay Tanwani |
| Location | |
| Category | Conferences - Seminars |
Abstract:
A long standing goal in artificial intelligence is to make robots seamlessly interact with humans in everyday life tasks. Robot learning from humans provides a promising route to bridge this gap. In contrast to direct trajectory learning from demonstrations, many problems arise in interactive robotic applications that require higher contextual level understanding of the environment. This requires learning invariant mappings in the demonstrations that can generalize across different environmental situations such as size, position, orientation of objects, viewpoint of the observer, etc.
In this talk, I will explain how we learn invariant representations with a family of hidden semi-Markov models for segmentation, recognition and synthesis of robot manipulation skills. The invariant segments (also termed as sub-goals, options or actions) adapt the movement of the robot in accordance with the external environmental situations such as size, position and orientation of the objects in the environment using a task-parameterized formulation. I will address learning these models in latent spaces to scale them with high-dimensional sensory data and discuss their Bayesian non-parametric extensions under small variance asymptotics for online learning with streaming demonstrations. Practical examples to perform remote manipulation tasks by teleoperation over satellite communication will be used to illustrate the performance of invariant task-parameterized generative mixture models.
Biography:
Ajay Tanwani is a PhD student in the Computer Science department at Idiap Research Institute and Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland. He is interested in robotics, machine learning and artificial intelligence paradigms for seamless human robot interaction in everyday life tasks. He recently finished his internship with Google X in Mountain View, California. Prior to doctoral studies, he completed his MS in European Masters on Advanced Robotics with Erasmus Mundus Scholarship from University of Genova and Warsaw University of Technology in 2009-2011. His noticeable recognitions include Best Manipulation Paper Award Finalist at ICRA 2016, World Summit Youth Award winner 2009, and winner of several all Pakistan robotics and engineering competitions (NERC’07, SOFTEC'08, COMPPEC'08, NASCON’08, IST’05).
A long standing goal in artificial intelligence is to make robots seamlessly interact with humans in everyday life tasks. Robot learning from humans provides a promising route to bridge this gap. In contrast to direct trajectory learning from demonstrations, many problems arise in interactive robotic applications that require higher contextual level understanding of the environment. This requires learning invariant mappings in the demonstrations that can generalize across different environmental situations such as size, position, orientation of objects, viewpoint of the observer, etc.
In this talk, I will explain how we learn invariant representations with a family of hidden semi-Markov models for segmentation, recognition and synthesis of robot manipulation skills. The invariant segments (also termed as sub-goals, options or actions) adapt the movement of the robot in accordance with the external environmental situations such as size, position and orientation of the objects in the environment using a task-parameterized formulation. I will address learning these models in latent spaces to scale them with high-dimensional sensory data and discuss their Bayesian non-parametric extensions under small variance asymptotics for online learning with streaming demonstrations. Practical examples to perform remote manipulation tasks by teleoperation over satellite communication will be used to illustrate the performance of invariant task-parameterized generative mixture models.
Biography:
Ajay Tanwani is a PhD student in the Computer Science department at Idiap Research Institute and Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland. He is interested in robotics, machine learning and artificial intelligence paradigms for seamless human robot interaction in everyday life tasks. He recently finished his internship with Google X in Mountain View, California. Prior to doctoral studies, he completed his MS in European Masters on Advanced Robotics with Erasmus Mundus Scholarship from University of Genova and Warsaw University of Technology in 2009-2011. His noticeable recognitions include Best Manipulation Paper Award Finalist at ICRA 2016, World Summit Youth Award winner 2009, and winner of several all Pakistan robotics and engineering competitions (NERC’07, SOFTEC'08, COMPPEC'08, NASCON’08, IST’05).
Practical information
- General public
- Free
Contact
- EDIC - [email protected]