Google suspended its robotics work for a while, but is back with a new idea: the Everyday Robot Project. Google points out that industrial robotics focuses on robots programmed to perform specific tasks in structured environments. This gives us machines that can build paper cups much faster and more accurately than any human being.
Robots programmed to perform specific tasks in unstructured environments aren’t usually so successful. This gives us robots that can fold one T shirt per minute and only with lots of help.
Google wants to do something different: robots that are programmed to learn new things in unstructured environments.
Sorting trash for recycling and other kinds of processing is like folding laundry: it involves unpredictable shapes and surfaces that may move around in surprising ways. The objects are not presented in optimized formats to make it easier for the robots to grasp them, and they require different actions in response to their different characteristics.
Neither sorting trash nor folding laundry is a task that human beings would mind giving up to robots. While these jobs are challenging for robots, they are excruciatingly dull for people. Automation of these jobs would be welcomed by the people who do them.
The same could be said for finding spills in grocery stores. Unfortunately, the robots that have been used in stores for this job can’t seem to tell the difference between a grape stem and a broken jar of pickles. They call humans to clean up the wrong kinds of messes and their work is not helpful.
So Google is working on robots which are programmed to learn things. The idea is that a combination of cutting-edge sensors and machine learning will allow these smart new robots to learn how to do daily tasks of many different kinds.
Google’s robots are practicing their learning skills on sorting trash right now. But they could, in the future, do things like washing produce… maybe even folding laundry.
Safety has to be a primary concern with industrial robots of any kind, but especially with robots intended to learn different kinds of tasks. The robots are likely to be collaborative robots. Engineers at Proctor & Gamble, where they’ve been working with collaborative robots formerly a decade, point out that robots can’t be judged safe without thinking about the entire application.
Take the new Rexroth robots that can perceive contact with a human being and stop. This is clearly much safer than a robot that completes its movement even if it touches a human coworker. But suppose the robot is moving a box from one place to another. If the box touches a human, will the robot stop or complete its movement? You might have a safe robot that can hit a coworker with a box.
Robots that can learn new behaviors definitely bring up these questions.
This is certainly a step forward for robotics. While we wait to see what happens next, let us be your first call when you need support or service for your Rexroth electric motion control. We are specialists, with decades of experience with Rexroth solutions.