Register for free to
view the full innovation profile.
Flexibility to pick up varied objects with one hand, able to pick up objects never encountered before, can handle fragile items.
About
What is the problem? Robotic hands are typically designed and made to perform a limited number of actions which have either been learnt or for which the hand is specifically designed. Current hands have a very limited ability to learn a grasp for a novel object which has not been encountered before. Our new Solution? A robotic hand which from one learned example can determine the optimal spatial relationship between each part of the hand and the surface contours of the object to maximize the likelihood of a grasp being effective. Benefits of the new technology Flexibility to pick up varied objects with one hand, able to pick up objects never encounter before, can handle fragile items. Background The ability to grasp a variety of novel objects is an open problem. Researchers have tackled this challenge and developed software to solve the problem of transferring grasps across objects. Given one or more grasp examples the algorithm aims to capture the configuration of the hand and its fingers with respect to neighbouring object surfaces and to adapt the hand and fingers to the shape of the novel object. The approach relies on a pair of models that capture two complimentary aspects of the grasping process. Firstly a contact model captures the configuration of one hand part relative to local features of the object surface. When given a 3D model of a novel object a contact model is developed which allows the robot to compute a set of poses for each and every hand part. If a robot is equipped with N hand parts up to N contact models can be instantiated. Secondly a hand configuration model captures a set of hand shapes that comply with the demonstrated grasp developed from the contact model. This allows a set of corresponding poses to be simultaneously realised by the hand parts without diverging from the demonstration grasp. The algorithm generates many possible grasps, each grasp having a set of hand part poses that independently comply with the contact models and the hand configuration model and selects the optimal grasp. This dual approach improves the adaptability of the hand compared with previous approaches.