University of Cincinnati Researchers Make Couch-Moving Robots

Recall a time you tried to move a couch with a friend or family member. Generally, such a task requires coordination and partnership (or teamwork if it’s a particularly sizable couch), pushing and pulling simultaneously in order to efficiently move the object. Cooperation is the key in order to move furniture without a hiccup, and it’s precisely why University of Cincinnati student Andrew Barth decided to use this simple task to test robot independence.

The experiment was conducted at the Intelligent Robotics and Autonomous Systems Lab at the University of Cincinnati under Ou Ma, an aerospace engineering professor.

A team of student researchers created an artificial intelligence that could train robots to move difficult objects, proving that they could work together. The idea was to move a couch, but for the experiment, the researchers made the robots move a long virtual rod through different obstacles. If they were able to move the object through those obstacles and out a virtual door, then the experiment would be successful.

A More Robust System

Moving a couch seems simple enough. If humans were to do it, we would probably devise a plan beforehand: scope out the room, identify difficult points, and come up with the best strategy to move the piece of furniture without delay. The moving process would also require verbal communication — cues like “1, 2, 3” or commands like “Push!” make the job easier.

But these UC doctorate students wanted to ensure that their algorithm was strong enough to withstand any potential problems, which is why they tested the robots’ cooperation skills with as little communication as possible among the machines. There were no plans devised among the two robots and no communication during the moving process.

Instead, the robots relied on fuzzy logic, an intelligent control technology that replicates how a person would make decisions when faced with situations that go beyond a clear binary. This means that robots have self-regulation in the same way humans do: they can identify mistakes and learn from them, adjusting their decisions as they go along. By replacing a simple yes-or-no binary with artificial intelligence that recognizes the degrees of truth between certain variables (like right or wrong), the robots are able to make sound choices in any given scenario.

This method paid off. The robots were successful 95% of the time in completing the task. They were able to move the rod or “couch” through two virtual obstacles and out the door.

But an even more important measurement was how well the robot partners navigated a completely new scenario without any re-programming or re-training. In this regard, the robots were able to complete the job 93% of the time in simulations. Even when other variables were altered, such as the size of the “couch,” the machines showed almost equal success.

“If you can train robots to work semi-independently with as little information as possible, then you made your system more robust to that failure and made it easier for large groups to collaborate,” says Andrew Barth, a UC College of Engineering and Applied Sciences Ph.D. student.

Barth has also published a study on this project for the Intelligent Service Robotics journal, co-authored by Professor Ma, Ph.D. student Yufeng Sun, and Lin Zhang, senior research associate.

Scalability

The idea behind these robot partners is to create scalable and sustainable artificial intelligence in almost any scenario.

“Ultimately, we want to expand this to 10 or more robots working cooperatively on a project,” Barth adds. “If you want to build a gigantic habitat in space, say, you’ll need a lot of robots working together. But if you were relying on a communications network and it goes down, then your whole project is done.”

Since the robots can also work semi-independently, they can continue the job even when one robot on the team malfunctions or comes across a problem.

The same is true for the reverse: whether you add two robots or four into the mix, they can quickly adapt and perform without the need for retraining, making the technology scalable, a must for any big project.

The larger vision is to create a team of robots that can work cooperatively. Nowadays, most robots are increasingly capable, but they also work alone. According to Professor Ma, the future needs to see robots working together as people do.