The idea of a hive or group mind where every individual is mentally connected is a mainstay of science fiction, such as the Borg from Star Trek or the more recent character Unity from an episode of the animated series Rick and Morty. The concept is generally inspired by naturally occurring self-organizing biological systems, such as ants and bees, that results in so-called swarm intelligence.
Individual ants are simple organisms when they are on their own, able to perform simple tasks, and only one task at a time: searching for food, laying eggs, or buildings walls out of soil. However, when you have a large number of ants together in a group, they form complex social networks that can perform complex tasks for the collective good of the group or colony.
As a collective intelligence the ants will allocate labor tasks, coordinate movement within the nest or across distances, and build detailed, elaborate nest structures. They can also organize to attack and overtake neighboring colonies. The swarm intelligence can also respond to external stimuli by changing their cooperative behavior, such as when ants will swarm around and possibly attack a predator—or inquisitive child—damages their nest.
This capability of swarm intelligence through power in numbers is what inspires many engineers and scientists to strive to recreate this phenomenon artificially. One such engineer is Michael Otte, a professor of aerospace engineering at the University of Maryland. He looks at natural swarm intelligence and applies it to robot decision making, combining the inspiration from insect colonies with the challenge of turning science fiction into reality.
“The concept of a ‘group mind,’ in which multiple consciousnesses are linked into a single intelligence, has been a plot device in science fiction literature since at least the 1930 novel Last and First Men,” said Otte. “I wanted to see if I could successfully apply that concept of a group mind in robotic swarms. Each robot has just a little bit of computational power, but together, they have a lot more than that.”
Wirelessly connecting a large number of robots, which are essentially artificial “brains,” into a single, complex computational entity makes what roboticists call a swarm. Otte trains swarms of these robots to connect to each other, communicate with each other to share data to form a detailed picture of their environment, and to collectively determine how to respond to it.
“In a robot swarm, each robot is programmed with simple rules. However, when robots interact with each other and their environment, complex behaviors emerge. Swarms of robots often use emergent behavior to accomplish a task; for example, in food aggregation, all robots pick up food, carry it around and drop it randomly. If we make the probability of dropping food proportional to the amount of food nearby, then most of the food will eventually aggregate in a single pile,” Otte said.
The goal is for Otte’s robots to combine both their physical and their computational resources in order to solve a common problem. His process involves hundreds of individual robots that merge their computing power to become a single distributed computer “mind.”
Otte’s research, recently published in The International Journal of Robotics Research, describes the methods he used to train a legion of Kilobots—simple robots that are only 3.3 centimeters tall—to accomplish something bigger than themselves.
Each Kilobot has a single light sensor, which is capable of discerning only a single light value. Put together a swarm of Kilobots, however, and they can combine their sensor data to ‘see’ across their entire environment. Each robot can communicate with other nearby Kilobots, bouncing wireless infrared signals off of the ground and up to neighbouring robots. This connection forms an artificial neural network across the swarm, which enables the swarm to detect and recognize images created by projected visible light.
The resulting computational entity is what’s known as an artificial group mind.
The robotic swarm uses it’s sensor data to collectively “see” across the environment, and the group mind determines what’s occurring in the environment and decides which behavioral response the swarm of bots should perform. For example, Otte trained the group mind to recognize and differentiate between a peace symbol and a biohazard symbol projected onto the floor. The group mind creates a smiley face if it recognizes a peace symbol, and forms a frowny face if a biohazard symbol is recognized instead.
“The group mind is trained to recognize the important environmental states that we might encounter, and an appropriate emergent behavior is prescribed as the desired response for each state. For example, given the distributed light sensor data that the swarm sees, the group mind can be trained to recognize if a peace symbol or a biohazard symbol is projected on to the environment,” Otte explained. “Some emergent behaviors require different groups of robots to perform different actions; furthermore, the swarm may be divided in different ways depending on which emergent behavior we desire. In this case, if the group mind recognizes a biohazard symbol, then it responds with an emergent frowny face. If a peace symbol is recognized, then the swarm creates a smiley face instead.”
Otte trained his Kilobot swarm using an algorithm with a tried-and-true set of rules commonly used in artificial neural network research. What’s new and interesting about his work, however, is that the algorithm was modified to be successfully applied across a distributed swarm of many robots connected by a wireless network.
“Wireless communication is inherently unreliable; messages can be dropped between robots, and individuals can fall behind the group,” explained Otte. “We accounted for this by programming neurons within the network to wait for neighbors who had fallen behind to catch back up to where they should be. This strengthens every robot’s neural pathways at the same rate over time.”
The ability to do this is especially useful for swarms such as Otte’s, with different robots programmed to perform different actions. Each robot is location-dependent, which means it determines its role in the collective response behavior, such as whether to move to form part of an eye or part of the smile, by where it starts in relation to its neighbors. If a robot falls too far behind its neighbor, it could compromise the response action, which is defined by the coordinated movement of the whole.
In other words, one rogue robot could spoil the whole swarm.
“In a sense, the robotic swarm is only as strong as its weakest individual. We need for all of the neurons in the entire brain to learn as a group. By waiting for individual robots who have dropped messages due to the unreliability of wireless networks, the swarm learns more efficiently, making itself stronger in the long run,” says Otte.
There are many different ways to design, train and test robotic swarms, and developing these technologies into useful autonomously operating systems remains a chief aspiration for many engineers and roboticists. In the end, what kind of application could these robot swarms see?
Otte believes there will be particularly high value in deploying robot swarms in the face of an unknown or hazardous challenge, such as search-and-rescue missions after a catastrophic natural disaster, where establishing risk and response need is an essential first step.
Interested in more on swarm robotics? Check out these stories: