english only
Dynamical Control
The aim of this research is to give a robot the ability to execute actions in a robust way. More
precisely, the robot performaing an action should adapt to sudden changes in the environment, or
to external forces acting on its limbs.
Traditional approaches, which make a clear distinction between action planning and execution,
fail to produce such a robustness. The dynamical system approach to robot control overcomes this
limitation because action is specified by a dynamical system acting on relevant variables. Thus, the
system continuously adapts to external changes or perturbation. We take advantage of the attractor
properties of the dynamical system to ensure that the desired final state will be reached.
Our research focus on two questions. The first question is how demonstrations can be used to
modulate a dynamical system specifying an action, in order to make it look more like the demonstrations.
The second question is how it is possible to combine multiple dynamical systems acting on different yet redundant variables.
Those two question are key to the developpement of imitation abilities in a dynamical system framework. The second question
relates to the choice of variables for the dynamical system, and the first question relates to the design of the dynamical system.
This research is partially funded by the RobotCub European
Integrated Project.
Videos
|
|
|
Dynamical control of reaching movement, 10.6Mb (avi) |
Imitation of reaching movement, 12.4Mb (avi) |
Selected publications
People involved in this project