The July 2022 issue of IEEE Spectrum is here!

Close bar

Lingodroid Robots Invent New Words for Time

These robots are inventing their own human-like language to describe spatial and temporal concepts

3 min read
Lingodroid Robots Invent New Words for Time

Last year, we were introduced to Lingodroids, which are small robots capable of developing their own language. This isn't a computer language, but more of a human language, with words that we humans could speak if we wanted to. These words have been invented by the robots themselves, using a variety of games to establish correlations between specific words and places, directions, and distances. And last week, Scott Heath from the University of Queensland in Australia presented a new paper on how the Lingodroids have been teaching themselves brand new words for different lengths of time.

Space and time are important concepts for robots and humans alike, since they're the four dimensions that define most of our existence. So when robots want to meet, it's not enough to simply decide on where to meet, they also have to decide when to meet. 

Last year, we learned how the Lingodroids were able to use interactive games to establish words for places, directions, and distances. For details, read our article from May of 2011, but basically the robots randomly generate names for unfamiliar areas, and then share these names with each other every time they meet, using games to reinforce associations and define boundaries. By extrapolating on this method, the Lingodroids can also develop terms for directions and distances, and cooperatively create maps of places that they've never been to. From last year's paper, here are the location maps and and distance and direction terms developed from scratch by two Lingodroids:

So for example, using the above lexicon, two Lingodroids located at "jaya" (in the center of the map) have defined the location "mira" (which they can't actually visit) by agreeing that it's at "puga puru" (northwest a long distance).

The method for inventing new words to describe time is fundamentally similar to how the robots agree on words to describe locations and directions and distances: through conversational games. For example, the robots played a game called "how-long-since-we-met," where they'd randomly explore the maze and give names to the amount of time since they last met. This established a basic lexicon, which the robots would reinforce with other games, like "meet-at," where the robots would choose a place and time to, you know, meet.

The graph below shows the duration lexicons that the two robots developed; the x-axis is time (in seconds) and the y-axis is a measure of the individual robot's confidence in that particular word/duration association:  

By combining these two lexicons, the Lingodroids can have relatively complex spatial and temporal goal-oriented conversations, like "ropi huzu jaya fohu," which would mean something like "let's meet just a little bit north of the 'jaya' location about 35 seconds from now."

Over the last year, the researchers upgraded the robot hardware to little iRats (pictured at the top of this post), and their environment is now a maze shaped like an intertwined "U" and "Q" (for University of Queensland):

Next, the researchers plan to start working on more conceptual units of time that humans tend to use based on context, like "soon" and "later." Also, humans like to assign words to specific points in time like "noon," which is tricky because these words can refer to points that will never happen again, as well as points that occur cyclically. The overall idea is that getting robots to understand a full set of spatial and temporal concepts in a more human way will enable them to interact that much more effectively with us in a wide range of contexts, without requiring us all to learn how to speak robot.

Lingodroids: Learning Terms for Time, by Scott Heath, Ruth Schulz, David Ball, and Janet Wiles, was presented last week at the 2012 IEEE International Conference on Robotics and Automation in St. Paul, Minn.

[ Lingodroids ]

The Conversation (0)

How the U.S. Army Is Turning Robots Into Team Players

Engineers battle the limits of deep learning for battlefield bots

11 min read
Robot with threads near a fallen branch

RoMan, the Army Research Laboratory's robotic manipulator, considers the best way to grasp and move a tree branch at the Adelphi Laboratory Center, in Maryland.

Evan Ackerman
LightGreen

“I should probably not be standing this close," I think to myself, as the robot slowly approaches a large tree branch on the floor in front of me. It's not the size of the branch that makes me nervous—it's that the robot is operating autonomously, and that while I know what it's supposed to do, I'm not entirely sure what it will do. If everything works the way the roboticists at the U.S. Army Research Laboratory (ARL) in Adelphi, Md., expect, the robot will identify the branch, grasp it, and drag it out of the way. These folks know what they're doing, but I've spent enough time around robots that I take a small step backwards anyway.

This article is part of our special report on AI, “The Great AI Reckoning.”

The robot, named RoMan, for Robotic Manipulator, is about the size of a large lawn mower, with a tracked base that helps it handle most kinds of terrain. At the front, it has a squat torso equipped with cameras and depth sensors, as well as a pair of arms that were harvested from a prototype disaster-response robot originally developed at NASA's Jet Propulsion Laboratory for a DARPA robotics competition. RoMan's job today is roadway clearing, a multistep task that ARL wants the robot to complete as autonomously as possible. Instead of instructing the robot to grasp specific objects in specific ways and move them to specific places, the operators tell RoMan to "go clear a path." It's then up to the robot to make all the decisions necessary to achieve that objective.

Keep Reading ↓Show less