Simplifying grasping complexity through generalization of kinaesthetically learned synergies

There has been a growing enthusiasm to use anthropomorphic hands of humanoid robots to manipulate every-day objects and tools designed for humans. However, multi-fingered grasping imposes a formidable control challenge due to the high dimensionality of the joint space and the difficulty to form a functional grip on objects. We propose a hybrid technique based on grasping synergies extracted from kinaesthetic demonstrations on a given object with a primitive geometry - a cuboid in this case - and passive kinematic enveloping as a generalization technique. Experiments were carried out on an iCub humanoid robot using everyday objects such as a telephone receiver, a computer mouse, three white board markers bundled together, a fencing handle, a compact disc keep case, and a drinking glass. We prove that the primitives extracted from kinaesthetic demonstrations on a cuboid can be generalized across a majority of the above real world objects.