This robot can help you meet the requirements of your online shopping
Imagine the moment you have fingertip vacuum cleaners — unless you use hallucinogens — then you shouldn’t imagine that. Each absorber comes in a different size and flexibility, making one finger ideal for sticking to a flat surface like cardboard, another more suitable for round things like a ball, another better for something irregular, like a vase. By itself, each figure can be limited to what things it can handle. But, together, they can work in a team to manipulate different objects.
This is the idea of Ambi Robotics, a lab-grown startup that is now emerging from stealth mode with an operating system to launch robotic classifiers and such manipulating machines. The founders of the company want to put the robots to work in jobs that should be frightened by any rational machine: Collecting objects in warehouses. What reaches people so easily – catching any object that isn’t very heavy – is really a nightmare for robots. After research in robotics labs around the world, machines are still not close to our skill. But maybe what they need are finger tip vacuum cleaners.
Ambi Robotics was founded at the University of California, Berkeley A research project called Dex-Net which models how robots should hold ordinary objects. Think of it as a version of robotics that AI scientists build to perceive images. To train machines to detect a cat, researchers need to build a database of many, many images of felines. Each time, they would draw a cat around the cat to teach the neural network: Look, this is a cat here. After the network looked at a number of examples, he was able to “generalize” them to a new image he would never have seen if the cat was automatically detected.
Dex-Net works the same way, but for robots. Working in a simulated space, scientists create 3D models of all kinds of objects, and then calculate where a robot should touch each to get a “strong” handle. For example, if you wanted to catch robots on a ball around the equator, don’t try to pinch one of the poles. That’s obvious, but robots need to learn these things from scratch. “In our case, the examples are not images, but ones that have strong points for capturing 3D objects,” says UC Berkeley robotist Ken Goldberg, who developed Dex-Net and co-founded Ambi Robotics. “Then when we got into that network, it had a similar effect. It started to generalize to new objects.” Even if the robot had never seen a particular object, it could have been called to train with one of its galaxies other to calculate how to better understand objects.
Consider a grotesque ceramic coffee cup made in art class in elementary school. Maybe you chose to adapt in an absurd way, but you probably remembered to give it a handle. When you handed it to your parents and they seemed to like it, they grabbed it by the handle; they saw the part they had in professionally manufactured coffee cups and so knew how to get there. Ambi Robotics ’robotic operating system, AmbiOS, is the equivalent of that previous experience, only for robots.
“Humans, we are able to deduce how to deal with this object, even if it is not like a glass that has ever been made,” says Stephen McKinley, co-founder of Ambi Robotics. “The system can reason what that other object is like, so if you select that part, you can reasonably think it’s a decent schema.”