Training robots to strategies to sort and sift through our disorganized stuff
As much as I’d love for a robot, or really, anyone, to come through my kids’ room to clean up, that kind of work is actually really difficult to get a machine to take care of. It’s not that robots lack motivation like my kids, but instead would literally be overwhelmed by the existence of clutter everywhere. Programming a machine to put objects in their proper place isn’t too hard, but helping them make sense of everything else piled around a target object can be a much more difficult feat.
Dealing with disorder
In a controlled environment, robots excel at what is referred to as “Pick and Place” (P&P) activities. When you can be sure the targets will be consistently discoverable for the robot, such as when they’re part of a factory conveyor belt, robots can beat humans with the amount of precision they have when placing that object somewhere else. This makes sense, because humans have evolved for mostly uncontrolled environments, and if we lack any precision, we make up for it with flexibility.
Outside of a factory, imagine you’re reaching to grab some sriracha sauce that has bizarrely been pushed to the back of your refrigerator. You barely have to think about reaching into that space and quickly selecting the object you’re looking for, as well as adjusting the objects around it that may get in the way of its removal. A wobble here and a nudge there is often all you need in order to do something that most robots would struggle with. Thankfully, researchers would like robots to one day be able to fetch us our hot sauce, and so they’re working on programming that lets robots cope with cluttered environments like a refrigerator.
Testing and training
One such robot is the Home Exploring Robot Butler, or HERB. As the name implies, HERB’s programming allows it to try out new solutions to these types of programming, and its creators have already seen it invent some novel solutions to dealing with cluttered environments. In one case, the robot used it’s highly-flexible wrist joints to hold an object in the crook of its arm as it moved it. It can come of with these solutions thanks to being programmed with some basic understanding of physics in order to predict how an observed object will be likely to move, hopefully avoiding knocking too many bystanders over on when picking something up. To help keep accidents to a minimum, HERB can also prioritize objects that have been flagged as especially fragile or delicate.
HERB’s next hurdle is to work on follow-through. Its decision-making is rather final at this point, and once it commits to an action it has trouble making adjustments mid-movement, another refinement that our brains basically let us take for granted. Solving these problems may yield some very big gains, even beyond my kids’ messes or my need for hot sauce. Other robots venturing into uncharted environments, such as the KREX robot slated to one day explore Mars, will hopefully take advantage of this kind of creative, clutter-dodging problem solving as these methods continue to be improved.
Source: Robots get creative to cut through clutter, Scienmag
