Friday Feb 13th at 11.00 at IDSIA
Prof. Dr. Jonas Buchli
Agile and Dexterous Robotics Lab
Institute of Robotics and Intelligent Systems
Unlocking the treasure trove: Learning impedance control
Autonomous robots are about to step out of the labs and factory floors.
Out there they have to deal with an uncertain, complex and dynamically
changing world. Allowing robots to make sense and to behave robustly in
such a world requires us to combine control theory with machine learning
in a principled way. While impedance control, be that active, passive or
a combination, is long recognized as a principled and powerful way to
control robots, its use is in complex scenarios is still fairly limited.
In this talk, I will sketch out the tremendous opportunities that
active, variable impedance control offers when combined with model based
and learning control. By understanding that this combination is key to
versatile and robust robot performance, we will understand why active,
variable impedance control has been a bit of a forgotten treasure to
this date: its use requires solving a very hard nonlinear optimal
control problem. This problem can successfully be addressed with
learning and iterative optimal control methods.