Learning Robots
Some hardwired, pre-programmed robots perform impressive tasks but do not learn like babies do. Unfortunately, traditional reinforcement learning algorithms are limited to simple reactive behavior and do not work well for realistic robots. To learn in realistic environments, robots must use novel algorithms for learning to identify important events in the stream of sensory inputs, and to temporarily memorize them in adaptive, dynamic, internal states until the memories can help to compute proper control actions. With our collaborators at CSEM we are working on attentive sensing and on learning hierarchical control strategies. We are studying not only real robots but also virtual ones, living in 3-dimensional video game-like worlds with rather realistic simulated physics.
People Bram Bakker
Viktor Zhumatiy
Juergen Schmidhuber



Copyright © 2003 - 2012 IDSIA. All Rights Reserved.