Among the different types of collaboration we can envision between
human and robots in an industrial context, the teaching of assembly
tasks is very appealing in the sense that it could drastically reduce
the complexity of configuration of a robotic setup for a given
task. Furthermore, comparing humans and robots, it is obvious that
humans are very skilled in learning and reproducing a new task, able
to optimize their action to do faster and better any manipulation,. So
that humans should definitely be a source of inspiration for the next
generation of collaborative robots. In that context, we have been
looking at the kinetic and kinematic behaviour of humans while
performing assembly tasks, and in particular we compared the human
strategies in unimanual and bimanual tasks, to get first insights on
how this affects their efficiency. Going back to the knowledge
transfer to the robot, we investigated how a human demonstration can
be taken on-board to correctly adjust the set of generic manipulation
skills the robot may be built with. First, we looked at how a
demonstration can be segmented in a set of known skills, and how each
of them can then be characterized to provide needed information to the
robotic system. This presentation will briefly cover these different
points.

Keywords: human manipulation, teaching robotic assembly, action segmentation.