Robot Plays Fetch

One of the things you’ll notice about robot innovations if you follow the news is that they don’t necessarily work. You’ll notice that about Boston Dynamics’ Spot, who can kind of play fetch.

Here’s how it works. Take a dog toy and photograph it 400 times or more in a specific environment, so you can be sure of getting a snap of the toy from every possible angle. This will be enough for Spot to perform its machine learning task of learning to recognize the toy.

Now you can throw the toy, being sure not to send it out of the environment Spot knows. Spot will slowly amble over to pick it up, then come fairly close to you and drop it.

To play with a new toy or in a new environment, just go through the same process again, 400 photos and all.

Proof of concept

This is a proof of concept situation, demonstrating that it is possible for a robot to learn how to identify an object and then to do something with it. It’s a step toward building a robot that can pluck a bad apple off a conveyor belt.

ut italso demonstrates that robots can’t really play fetch very well. Or sort apples, either. Someday maybe they will be able to. Right now they can’t match the skills of the average three year old in either area.

Things robots are good at

With a good motion control system, a robot can move a specific object from one specific place to another specific place, very fast and very consistently.

It turns out that this is a very useful skill in industrial situations.

So if your Indramat motion control system can get your machinery to do its part in your facility, don’t feel bad that it can’t play fetch. Some things can be done best by robots, and some things — at least right now — can be done best by humans. Or dogs.

If you need support for your Indramat motion control, contact us immediately. We offer factory repair and reman for legacy parts.