The SARAFun project goes beyond traditional industrial assembly
applications where uni-manual robotic setups are considered by
employing dual arm human-sized robots to perform assembly tasks that
are typically executed by humans, in dynamic environments originally
designed for human use. In this talk I will present a dual arm robot
control strategy for assembly that can easily accommodate geometric
uncertainty typically arising in environments where humans and robots
coexist. The framework incorporates coordinated task space frameworks,
a simple 2-DoF articulated object model that can capture a folding
assembly task and a kinesthetic perception module based on external
force/torque measurements and proprioception. The latter is able to
identify in real time the DoFs while estimating the position of the
joints. Experiments with a "toy" articulated object as well its
assembly analogous parts will be demonstrated to showcase how the dual
arm control strategy has been used within the SARAFun project to
successfully assemble components via folding.