on January 22, 2015 at 2:00 pm
A geometric look at anthropomorphic action
From a purely mechanical perspective, an anthropomorphic system (a humanoid man or robot) is a system that is both redundant and underoperated. It is redundant compared to almost all the actions it has to perform because it has a large number of degrees of freedom (about thirty engines for a humanoid robot, more than six hundred muscles for a human). For example, there are several ways to take the same object. It is under-powered because it does not have an engine that would allow it to move directly from one place to another. To move, it must play on its posture and operate its limbs in a cyclical way; this is the role of locomotion. Acting on the world thus requires the combination of the two fundamental motor functions: displacement and manipulation.
Considering that a movement is a continuous function of time in space, that its image is a path, that actions are compositions of movements, the image of an action in space is a path. The reasoning is for the physical space, the space of the system configurations, the motor space and the sensory space. Thus any action, however complex it may be, is summarized in a simple geometric object: a path. The correspondence is of course not bi-univocal: any path is not the image of an action. It remains that the study of this correspondence makes it possible to bring a particular look at the computational foundations of anthropomorphic action. It is in this perspective that the different techniques of segmentation and movement generation, as well as the associated optimization principles, make it possible to better understand the structure of seemingly complex actions.
français
English