Teaching Robots to Drink, Relax and Play Catch
Event details
| Date | 19.01.2012 |
| Hour | 17:15 |
| Speaker | Prof. Aude Billard |
| Location |
polydome
|
| Category | Inaugural lectures - Honorary Lecture |
Teaching robots to do skills is a more challenging task than it may appear at a first glance, even when these robots resemble us. In this presentation, I will attempt to shed some light on some of the major challenges we face when transferring skills from humans to robots and will review recent works we did at LASA to allow robots to learn tasks that require careful control of both the kinematic and dynamics of motion. The algorithms are generic and can hence be ported to several types of robots, from classical industrial robotic arms, to multi-degrees of freedom humanoid robots. I will also emphasize the fact that learning is not just play and record. A key component of learning and adaptation resides in the ability to generalize. In imitation learning, generalization must come from very few examples to ensure that the duration of the training phase is bearable to the human teacher. Moreover, most approaches discard erroneous training examples, hence forcing the human teacher to be very skilled at the task. There is an advantage to include errors in the training to ensure robustness and better generalization. I will hence conclude this presentation by discussing some promising avenues in learning from failed demonstrations.
Links
Practical information
- General public
- Free