Now AI Can Help Humans & Robots Walk More Quietly

A new system has been developed that might make it more difficult for robots to get lost while they’re doing research. While some robots can grasp things and blow things up, they tend to lose track of their original location.

The system used in the new study uses a computer that crowdsources GPS coordinates with all the other computers around it. The system isn’t necessarily for navigation purposes, though – instead, it could help AI in that field be more accurate and aware of the bot’s precise location.

"We need to find a way to reduce errors in navigation because even if we plan the way to do our work with solid data, we still need to transmit back to the software what the robot thinks it’s supposed to do,” professor Minkeun Kim said in a statement.

Human beings do everything with perceptual estimates – moving cars, swimming toward a pool and singing with perfect pitch. But robots require precise data like GPS, radar and even human vision.

The new study explained how to use a computer-based system to make robo-agents highly accurate at navigating those situations. 140 robots were asked to choose the same angle and direction in a workplace while performing the same tasks. Then they downloaded these coordinates to a portable hard drive and asked the human engineers and AI operators to interpret the coordinates by using scanners and touch screens.

Now, when the technology is in place in a lab, the computer will take that data to a software that can calculate the right distance to every task and edge. Then, when it’s time to move, the remote operators and operators-in-training, while controlling the robots, will take the coordinates and create a layout for the robot.

In Kim’s lab, the robots that have this technology will manipulate small tasks and avoid getting lost in the process, the team has already trained the robots so they have the upper hand. While driving the rover, for example, the bot can either steer the rover into the center of the circle or follow the circle itself. If the bot went in straight, it would run out of road before it could connect to a map and find its original coordinates. If the rover went in the circle, then the bot would work to get to its original coordinates.

“The real beauty of this system is that it generates and displays a map that’s way more precise than the previous methods,” Kim said. “It gives us an opportunity to change the system for very specific tasks. If we need to grab something and drag it, it can stay closer to the center of the circle instead of trying to figure out where the objects are around it and making a guess.”

The system also could be immediately assembled in a given work environment, and partly responsive to different hands, where the robot will direct itself if a hand is next to it or not. Though this new system is still a rough prototype and only works at a very small scale, we hope to apply this technology to place robots near real-world events, like rescues, aid flights or even first-responder role-plays.

This system has the potential to be very meaningful. There are all kinds of sophisticated robots out there right now, but not very many of them are designed for quickly adjusting to new contexts. It could be used for that and can ultimately build the kind of flexible systems that will allow us to achieve the kind of precision in robotics that we expect in the future.