We cover all kinds of modular robotics around here, and when we do, we’re almost always talking about one overall robotic system made up of many different modules, some number of which can be individually controlled or swapped around. What these systems generally have in common is that there’s one brain (usually a computer sitting on a desk somewhere) that interprets all of the sensory data from the modules, and then provides directions to each module. Essentially, the individual robots form a nervous system that passes information to the centralized brain, which is the same way that humans work, and so do most non-modular robots.
While this sort of system works quite well in a research environment, the ideal use case for modular robots is to make them more decentralized, such that any individual module can be part of a nervous system or a brain on-demand, depending on what the robot as a whole is trying to accomplish. In a recent paper in Nature Communications, Nithin Mathews, Anders Lyhne Christensen, Rehan O’Grady, Francesco Mondada, and Marco Dorigo from universities in Lisbon, Brussels, and Switzerland, present the idea of a “mergeable nervous systems for robots,” with a framework for fully modular robotic systems:
We present robots whose bodies and control systems can merge to form entirely new robots that retain full sensorimotor control. Our control paradigm enables robots to exhibit properties that go beyond those of any existing machine or of any biological organism: the robots we present can merge to form larger bodies with a single centralized controller, split into separate bodies with independent controllers, and self-heal by removing or replacing malfunctioning body parts. This work takes us closer to robots that can autonomously change their size, form and function.
The robots used in this research are Swarmanoids, modular robots that we’ve covered extensively in the past, although in the paper, they’re referred to as “mergeable nervous system (MNS) robots.” A single MNS robot can consist of an arbitrary number of Swarmanoid units, and MNS robots can split themselves up or combine several robots into one robot, depending on their objective. Each MNS robot has one module that functions as a brain, but the brain module isn’t restricted to that particular module, and it can move around or split or combine, along with the physical structure of the robot.
Three autonomous MNS robots with LEDs illuminated in different colors.Image: Marco Dorigo and Nithin Mathews/Nature Communications
With one module acting as the brain, the rest of the robot modules (whatever their configuration) form the rest of the nervous system. Modules farther from the brain collect sensor data, passing it to their parent modules, which synthesize it and pass refined data up the chain. For its part, the brain module makes all of the high level decisions, which are passed back down to individual actuator modules.
The upshot of this ability is that MNS robots are extraordinarily flexible and resilient, in a way that most modular robots aren’t. You don’t have to worry about the vulnerability a single brain module, and given enough generalized modules, you can reform your robot swarm into as many configurations as you need while easily compensating for any physical damage that may occur.
MNS robots thus constitute a new class of robots with capabilities beyond those of any existing machine or biological organism: An MNS robot can split into separate autonomous robots each with an independent brain unit, absorb robotic units with different capabilities into its body, and self-heal by removing or replacing malfunctioning body parts—including a malfunctioning brain unit.
For more details, we spoke with Nithin Mathews, who, in addition to being first author on this paper, is currently a PhD candidate at Université Libre de Bruxelles and a software engineer at Netcetera, a software company based in Switzerland.
IEEE Spectrum: What is a robotic nervous system, and how does it work on traditional robots?
Nithin Mathews: The sensorimotor system that physically connects a robot’s central processing unit to its sensors and actuators can be seen as a robotic nervous system—quite similar (at a conceptual level at least) to a biological nervous system of an higher order animal with a single brain controlling its host body. In most of today’s robots, the structure of its nervous system is specified at design time and remain static through the robot’s lifetime.
How are robots with mergeable nervous systems different, and what advantages do they have?
Imagine being in physical contact with someone and–given mutual consent–being able to use their body as an extension of your own. You would see what they see, hear what they hear, and move their limbs as if they were your own. Such a feat is impossible not only for humans, but for all natural organisms. In science fiction, however, the concept of merging bodies is not at all new.
The best known example may be from the 2009 James Cameron movie, Avatar. In this movie, a horse-like creature (named direhorse) becomes an extension of its Na’vi rider’s body once he or she physically connects both nervous systems using protruding neural extensions available to both bodies. This neural bond can be used by the rider to effortlessly command the mount by communicating motor commands directly from his or her brain. Conversely, sensory information including pain felt by the animal become instantly available to the rider. In a nutshell, this is what we have tried to replicate in the robotic world.
The advantages are numerous: The control of an MNS robot needs no adaption during runtime even when the physical structure of the robot changes (to cope with new tasks, for instance). An MNS robot can also spontaneously borrow physical features only available to a peer robot long after its initial deployment—similar to a Na’vi rider borrowing the locomotion capabilities of a direhorse in the Avatar movie. Another very useful feature we were able to demonstrate is the ability of an MNS to detect and respond to hardware failures by invoking a self-healing procedure. Theoretically, as long as spare units and sufficient energy are available, MNS robot may not be prone to malfunctioning software or hardware components at all.
Can you describe what happens when one MNS robot splits, or when two MNS robots merge?
When splitting occurs, meaning when an existing body is separated into one or more separate parts, MNS robots are capable of instantaneously creating new brain units in each of those new parts allowing them to operate as new, fully functional robots with a nervous system adapted to the new body.
When two MNS robots merge, two brain units or two decision-making entities need to be integrated into a single body. Imagine holding hands with four of your friends and trying to run towards a moving target. This will be messy by nature as there are four separate entities making decisions that need to be coordinated. MNS robots, on the other hand, are capable of solving this issue at a technical level by seamlessly ceding control to a single brain unit and retrieving the control when necessary.
How could mergeable nervous systems be applied to a future generation of practical robots, and what kinds of tasks will they be useful for?
Mergeable nervous systems for robots have the potential to go a long way in the future. Applications in which robots need to physically cooperate with each other represent ideal candidates. This could be nano-robots operating inside the human body to robots operating on foreign planets (imagine an advanced version of a humanoid robot operating on Mars unifying its body and sensorimotor coordination with a rover it just hopped into).
What are you working on next?
Our focus is surely on moving from a two-dimensional planar surface to operating in a three-dimensional world. For this purpose, we may need to consider other ways in which physical connections can be made between individual robots. In this context, we will also be looking into other forms of locomotion—think of independent robots joining forces in a three-dimensional space to then perform a bipedal walk or executing a crawling, climbing or rolling behavior to overcome obstacles in their environment as it best fits.