Robots tend to be controlled by sensors and actuators connected to a central processing unit – a sort of robotic nervous systems. Traditionally, flexibility in the system is limited as these nervous systems are mapped strictly to the shape of the robot. The development of a more modular system formed instead by multiple units would allow more adaptability. Indeed, in principle robots could display more lifetime morphological adaptation than natural organisms, with robots of different capabilities, shapes, and sizes, configuring themselves as required.
However, limits to the predefined shapes that units can form into as well as the reliance on distributed control, has curtailed the coordination and control of modular robots, which have only been able to display a limited range of hardwired behaviours. The team behind the EU-funded E-SWARM project have recently reported that they have successfully designed a modular system that can adapt their arrangement, autonomously forming shapes and sizes depending on the task or environment.
Introducing mergeable nervous system (MNS) robots
The study recently published in Nature Communications
points out that the behavioural of most current modular robots functions under control paradigms akin to the biochemical signalling used by simple natural organisms, such as unicellular slime mould, which can alter their body composition. However, as is the case with these biological counterparts, they lack a nervous system that can holistically unite disparate parts into a functioning and adaptive whole. This means that while they can be individually autonomous, they rely on distributed approaches for coordination.
The E-SWARM team outlines how they created robots whose bodies and control systems formed new robotic systems as necessary, while retaining full sensorimotor control, regardless of shape and size. Within this MNS robotic system, where the units connected via the robot nervous system, the centralised decision-making unit is known as the ‘brain unit’. These mergeable nervous system (MNS) robots were able to merge by absorbing units of different capabilities into its body forming larger clusters with a single centralised controller; split into separate bodies with independent ‘brain units’, and self-heal by removing or replacing malfunctioning body parts including a malfunctioning brain unit.
The team set-up an experiment for ten robotic units. The units formed a series of MNS robots of different shapes and sizes, as they adhered to pre-determined behavioural rules. For example, the MNS robots all displayed the same coordinated sensorimotor reaction to a provided stimulus which involved ‘pointing’ at the stimulus, using light emitting diodes (LEDs), as well as retreating from it, when it was sufficiently close. When a cluster of units points to the stimulus, only the LEDs closest to the stimulus illuminate, regardless of the robotic unit to which they belong.
Building the agile robots of the future
Despite only engaging 10 units for the experiment, the paper’s authors point out that their system is designed to be scalable, both in terms of computational resources for robotic control and reaction time to stimulus, within the system. Looking to the future, the team suggest that robots will likely be designed for adaptability to changing task requirements and no longer only for specific tasks.
The project team are now looking to extend the concept to self-reconfigurable modular robots, beyond the realm of two dimensions into three, for example by with the design of flexible joints. The team suggest that advances in computational power and techniques should be able to compensate for the millions of years that evolution has been afforded to solve the same sort of design questions in nature.
For more information, visit:Project website