Robot See, Robot Do

Robots are great at performing repetitive task with precision and accuracy. They carry out specific movements without variation for hours, or days, on end. This is what makes industrial motion control and modern manufacturing possible. Robots aren’t good at everything, however. Machines are poor at adjusting to unexpected variables, and they have trouble finding ways to work around inconsistencies. They must be programmed to perform a task, otherwise they can’t perform the task.

There’s a lot of progress being made with artificial intelligence and machine learning, but teaching physical robots to manipulate the world around them remains a challenge. MIT researchers are working on a system that would allow robots to learn new tasks by through observation.

Planning with Uncertain Specifications

Machines follow a set of rules. If there’s a zero here do this, if there’s a one there do that. A black and white set of rules works well in a controlled environment where things are consistent and predictable. The real world is anything but predictable, however.

Humans don’t share this binary understanding of the world. We can recognize that an oddly shaped orange is still an orange. We solve problems and intuitively adjust when something unexpected happens, and we weigh our options when making difficult decisions.

Researchers from MIT want robots to consider multiple factors and potential outcomes to improve decision making and adaptability. The “Planning With Uncertain Specifications” system, “gives robots the humanlike planning ability to simultaneously weigh many ambiguous — and potentially contradictory — requirements to reach an end goal”. The goal is for robots to be able to learn how to do new tasks in a real world setting without specific programming.

Why? The average person could deploy a robot in their home, and teach it to perform a wide variety of tasks, without having any special training or knowledge of programming.

Great for home robots, bad for manufacturing

Domestic robots (that actually do a good job) would be fantastic. They would free our time for more worthwhile pursuits and improve our quality of life. We’re getting closer to a reality where home robots that can do our mundane domestic chores for us. We can pretty much automate vacuuming, mopping, and mowing the lawn, and researchers are working around the clock to crack the code on a laundry folding robot. A robot that could actually learn to reliably carry out new tasks by observing a human would be extremely beneficial.

It would be bad for manufacturing, though.

Industrial automation doesn’t need a versatile and thoughtful robot. It needs powerful, fast-moving, and consistent machinery. That’s what your Indramat motion control system delivers. Call 479-422-0390 for any of your Indramat service needs.