Hi, this is Wayne again with a topic “Robo-gripper grasps by reflex”.
[, MUSIC PLAYING ] Andrew SaLoutos, Historically, a lot of the interest in robotic manipulation has been focused on pick and place in a factory environment or in a more heavily structured environment.. But if we want robots to go into the home, we need to be able to deal with more unstructured environments and places where people live. And if you look at human manipulation, something that sets us apart, is our ability to react and use our local information at Our fingertips. And we don’t come up with these super complex manipulation plans ahead of time.. A common approach in other manipulation systems is to use vision and use cameras and plan all the details down to specific fingertip locations on an object.. That slows down the system and you depend heavily on how quickly you can get information from your cameras.. So what we’ve done is we still use a vision system.
And we have a camera, basically take an image of the scene and say here’s the closest object that you can grasp. And for the camera. That’S it.. We’Ve put proximity, sensors and contact sensors in the fingers.. So as it approaches the object, it can see the surfaces that it’s coming up against and react around them. And then, when it knows that something’s in its grasp, it can close and measure the forces that it sees so that it knows whether or not it’s successfully Grasped the object.
Hongmin Kim The cable driven mechanism is similar to our tendons in our fingers, in that the cable drives the joints., So it can flex or it can extend at the joints. As opposed to other systems having actuators on the joints. We use a cable driven mechanism to place all the actuators at the base, which makes the joint really low inertia, which reduces the weight at the fingers, making the fingers really fast and reactive.
Andrew SaLoutos. I guess the key way that our system is different is that if something goes wrong, so if the vision information is slightly off and the cup is not exactly where the camera thought it was, our system can react without having to consult the vision system again and Ask for new information or without ruining the entire manipulation plan., And we do that using only the local information on the fingertips.. A key part of our approach is that we don’t want to basically assemble a library of very individualized solutions for all of these objects.. We want to grow our capabilities, but in a way that we stay generalizable., We didn’t pick a single cup and design our controller for that cup.. We want to sort of grasp this large class of cylindrical objects and, as we add more, we want to keep that mindset.
.