students and robot in lobby
One of the robots navigating through the College of Engineering lobby.
Photo: Colleen Stepanian

Picture yourself walking down a hallway. You see another person approaching. Often, a combination of physical and verbal cues will dictate that you will move one way and they another. Sometimes, though, wires cross. You zig when you're supposed to zag, and you almost run into the other person.

What if that other person is actually a robot? This type of scenario and others are under review by Temple University Mechanical Engineering Assistant Professor Philip Dames (pictured at right). His work, funded by the Amazon Research Award program, is studying how certain robots can safely navigate through a crowded, dynamic environment.

"There's basically a big section of warehouses that are currently designated as robot only," Dr. Dames said. "To remove this separation, robots need to understand where people are now and to predict where the people might be going in the near future."

Dr. Dames also added that the results could apply beyond warehouse or manufacturing environments to an office or a retail store where there are likely to be more people than robots.

The project is reminiscent of other recent robotic and AI-driven consumer solutions. The grocery chain Giant introduced a robotic customer service supplement which travels around the store and alerts customers and staff to things like spills or trip hazards.

The project focuses on four main components: localization, object recognition, multi-target tracking and a conventional neural network (CNN) based controller. These will empower the robot to know its own position, to identify and track objects and to navigate fluidly through a crowded space without bumping into people in real-time.

"I'm trying to take pieces from these different sub-fields across robotics and beyond and apply them in a way to help the robot move more fluidly," Dames said.

Dames plans to work with a graduate student on the one-year project and utilize a pre-built robot chassis, adding that core concepts could also inform future Amazon fields such as Amazon Scout or Amazon Prime Air. It could also power broader, non-Amazon projects, such as the FedEx Same Day Bot. The software tools that Dr. Dames and his student develop will even be made open-source so that other researchers can utilize them in their own projects.

So, how close are we to seeing robots "working" in an Amazon fulfillment center, maneuvering through your local Whole Foods to restock kombucha, or even buzzing past you in the office? Dr. Dames expressed optimism about the success of the project, with an eye on the broader application.

One of the fun parts of research, Dr. Dames noted: "You really don't know what will happen until you try it."