Taking inspiration from biology
The multidisciplinary Swarm-Organ project investigates how systems containing large numbers of autonomous but relatively simple agents can collectively organise themselves into complex spatial arrangements despite each agent having only local awareness. In the field of ICT, such agents may be robots, and the collective system is often called a swarm. In swarm research, a number of biological systems have provided the main inspiration to date.
The main inspiration for this project comes from the embryonic development of the limb, a relatively unexplored system taken as an example of large-scale swarms of cells. The agents studied in developmental biology are individual cells, and the collective system is a multicellular tissue or organ. How this biological system manages to achieve such exquisite organization, spatially-controlled division of labour (muscle cells, nerve cells, bone cells all accurately positioned to function together), and robustness is still not well understood, but it is clear that much of this functionality is built into their genome as gene regulatory networks (GRNs). Swarm-Organ scientists are exploring the use of GRNs as a control method for both biological and technological systems.
The project has already achieved some important goals. Starting with an agreed standard 2D model of multi-agent swarms (the model is composed of two parts: a representation of gene regulatory networks and a representation of the physical agents and their movements), a new simulator software has been implemented. The main advantage is that the same program ran by the simulator can directly run on the chosen platform of real bots: a large swarm of low-cost mini-robots, the kilobots. The consequently limited capabilities of these bots impose several constraints on the driving algorithms and also on the physical properties. One of the issues is the lack of directional sensing of the bots. This fact actually brings the focus really down to the essential basis of the swarm behaviour. Examples of swarm behaviours have already been shown with kilobots, including the recent work from a Harvard group (demonstration with a swarm of 1,000 kilobots, able to make shapes that are predefined in the form of a binary bitmap).
Swarm-Organ researchers have successfully performed various tests of patterning and morphogenesis on the new simulator: generating stripes with GRNs, generating patterns with a reaction-diffusion mechanism (Turing type), shape formation, interaction between epithelium and mesenchymal tissue, cell sorting, tracking, etc. Some of the first basic patterning tests (morphogen gradient and stripe formation with a GRN) have been also translated into real-world demonstrations with kilobots. Initial results on morphogenetic self-organization without directional sensing and the use of GRN for adaptive swarm behaviour have recently been published.
Same simulation of a morphogen gradient in kilobots (left) and simulator (right).
Creation of a stripe pattern in kilobots using a Gene Regulatory Network (left). Shape formation in the simulator using morphogen gradients from 3 sources (right).
The work for the second half of the project will address different behaviours for healing and regeneration, environmental adaptation, tracking randomly-moving objects, surrounding them and finally herding them into a predefined shape, and several experiments on a 200+ kilobots swarm. The final result of the project will be a better understanding on how simple agents with identical instructions can achieve robust spatial organization and division of labour in a collective and self-adaptive manner.
The project is developed by a consortium of four partners from Spain, United Kingdom and the Netherlands. It started one and a half years ago and is one of the seven projects integrated to the Coordination Action FoCAS (Fundamentals of Collective Adaptive Systems).