For robots, the future is modular

“Mergeable nervous systems” can create flexible, self-adapting robots. Michael Lucy reports.

Three autonomous MNS robots.
Three autonomous MNS robots.
Marco Dorigo and Nithin Mathews

One mind, one body: most of the time it’s a pretty good rule of thumb, but new developments in robotics may make it obsolete. The idea of a hive mind or swarm intelligence is making its way from the pages of science fiction onto the laboratory floor.

In a paper published in Nature Communications, Marco Dorigo of the Free University of Brussels and colleagues describe a system of “mergeable nervous system” (MNS) robots that can combine and divide their bodies and control systems.

Most robots are designed with a “nervous system” similar in outline to your or mine: sensors deliver input to a central processing unit (a “brain”), which in turn issues instructions to move or perform other actions. Though modular robots that can work together do exist, their control systems generally consist of simple signalling between units – something like single-celled slime molds that can exhibit a degree of organised behaviour without central control.

Dorigo’s MNS robots, however – which use the marXbot hardware platform – can join together to form a single robot with a single nervous system in which one unit functions as the “brain” and the others as the body.

The mergeable nervous system concept. MNS robots consisting of a single robotic unit (center) self-assemble into larger MNS robots with a single brain unit (shown in red).
The mergeable nervous system concept. MNS robots consisting of a single robotic unit (center) self-assemble into larger MNS robots with a single brain unit (shown in red).
Dorigo et al. Nature Communications

This allows unprecedented flexibility: a robot can split into smaller autonomous robots, each with an independent brain unit; it can bring in new units with different capabilities to complete a task; and it can swap out damaged units, including a damaged brain.

The authors are not shy about what they see as the import of the development: “a new class of robots with capabilities beyond those of any existing machine or biological organism”.

The key element of the design is making the nervous system able to cope with a changing configuration of input sensors and body parts.

The nervous system of each unit contains a tree-like topological description of the physical arrangement of the robot as a whole with the brain as the root. In particular each unit has a representation of itself and all its “child” units, so that it is prepared to split off with its children and become the brain of a new robot at a moment’s notice. Sensory inputs and commands are passed up and down the tree structure.

Ten MNS robots react to a stimulus by pointing at it with their green LEDs, and retreating when the stimulus is too close (point and retreat behavior). The ten robots then merge to form two MNS robots, which display the same point and retreat behavior. The two MNS robots then merge into a single MNS robot which displays the same point and retreat behavior.
Dorigo et al. Nature Communications

At present, MNS robots that can change body shape and size while retaining fine sensorimotor coordination but they are limited by rigid connections between units and having to keep their wheels on the ground.

The next step, the researchers say, is to build in three dimensions and with flexible joints.

“In the future,” they write, “robots will not be designed and built for a particular task.”

Instead, they hope to build swarms of robots that can choose the right configuration for whatever jobs they are called on to do.

Michael Lucy is the online editor of Cosmos.
Latest Stories
MoreMore Articles