Grippers that conform to objects surfaces are really useful. Some have looked at hand designs as all manmade should be made for grasp of the human hand. but it’s not easy to teach human grasp to a robot as we do grasp objects differently, e.g. how one holds a pair of chopsticks. I have seen different people holding a pair of chopsticks differently but yet able to finish a meal with the grasp. Secondly, it’s easy for humans to know how we are holding an object but it’s difficult to determine the orientation of the object if we use a robotic hand to hold the object.
The 2/3 finger design is simple and it can work really well maybe 90% of the time, which is good enough. Velco 2G is a passively adaptive gripper. It’s gripper is able to conform to any object it grasp. This is similar to the Robotiq grippers as well. BarrettHand is another gripper that’s falls in this category but it’s better because 2 of the 3 fingers of the BarrettHand can rotate 2 of the fingers around the palm axis and that allows changing the grasping style of the gripper. I find this useful for some applications.
Having the Velo 2G on the PR2 is great. It’s improves the grasp of the PR2 and i hope to see some innovation outside of software that can reduce the computation power requirements of the robot. My belief is that we need to produce robo-humancentric objects to improve the way the robots is able to perform task in a human-centric environment.