In the Loop with Loopy

When it comes to repetitive work, robots can hang with the best of them. In fact, they are the best of them. On assembly lines and in warehouses, robots can tirelessly carry out the same set of tasks over and over again with precision and speed in a way that humans cannot. But before you start worrying about a robot takeover, you should know that robotic systems become very brittle when faced with unstructured, real-world environments. Under these conditions, they cannot simply carry out the same set of steps repeatedly. Rather, they must perceive their environment and learn how to adapt to dynamic conditions.

Traditionally, these features are implemented with a computer vision system, advanced sensors, machine learning algorithms, and a powerful computing unit. A lot of progress has been made using this basic approach, however, the technologies are very expensive. Furthermore, there are still many issues with robots failing to adapt to situations that are outside of the distribution of the training data.

Roboticists at West Virginia University have taken a different approach in an effort to solve these problems. Instead of relying on cutting-edge technologies, they are banking on emergent properties that arise from groups of smaller and simpler robots that work together — something like what is seen in the behavior of swarms of insects or the growth of a plant’s roots. Their ideas were implemented in a robot named Loopy that is composed of a number of interconnected robotic “cells.”

Unlike the traditional top-down model of robot control systems, Loopy is not programmed as a whole to carry out a specific task. Rather, each cell has independent actuators and simple sensors — like light and temperature sensors — as well as a sensor that determines the angle of the joint that connects it to other cells. These cells are given some basic directions about how to interact with the world using this onboard equipment.

A set of 36 cells were assembled to create Loopy. The researchers then wanted to understand how this robot might be able to locate and mark the boundary of an area contaminated by a toxic spill. So they set up a tabletop test environment and let the show begin. The team also threw lots of obstacles and different types of surfaces at Loopy, because they were especially interested in how their robot would respond to unexpected and dynamic conditions.

The work is still in the early stages, but the team is interested in seeing how Loopy deals with problems, and how that would compare with a robot that was fully programmed by humans. Initially, they note that some very unexpected and interesting solutions have been observed. They believe that observations like these will drive future investigations forward and ultimately move the state of the art in the field forward as well.