1. Introduction
Service robotics, as it is widely performed today, usually assumes a given service to be carried out by a single robot or, at most, by a small group of them working together. In any case, though, the concept of cooperation is intended more in the sense of a relay race, than in the sense of an actual team effort for achieving each single task. Each robot, in other words, is assumed to be able to cope with the basic problems of autonomy alone, i.e. locating itself, navigating within its environment, and in case also planning its own future actions.
A new and totally different way is the so-called swarm robotics that, as opposed to the more traditional approach, does not necessarily assume each robot as a stand alone independent unit. On the contrary, swarm robotics assumes that a given mission is the result of a joint action of a swarm of simple units. Such units, in theory, might even be unable to perform the bare locomotion without the aid of others of their kind.
This approach finds its theoretical roots in recent studies on swarm intelligence [1], i.e., in studies of selfassembling and self-organising capabilities shown by animals such as social insects [2]. With this sort of approach, cooperation becomes of capital importance for the success of an overall mission. Indeed, since there is no predefined role, an artificial swarm (which we label swarm-bot) can be, as its counterpart in Nature, extremely robust: the function initially endorsed by a failing unit would simply be replaced by a reorganization of the whole group. Clearly, such a characteristic implies a swarm to show an overall behaviour which is both adaptive and emergent. Adaptive because it needs to change itself opportunely in order to cope with the surrounding world, and emergent because each unit (which we label s-bot) has no global cognition of an assigned mission: they simply respond to the external stimuli with specific local predefined behaviours.
In this respect, exactly as simple behaviours interacting with each other would let a more complex one emerge, a group of s-bots acting on their own locality, would be able to perform an assigned task as the result of their team effort.
It is important to notice that physical reconfiguration, though, is not the only characteristic distinguishing a swarm-bot from a more traditional service robot. Selfassembly is the other important feature. Once a task is requested, in fact, a swarm-bot not only needs to evolve constantly its physical shape until the completion of the assignment is reached (final goal), but it also needs at the beginning physically to assemble its components (s-bots) from scratch. Moreover, once the goal is achieved, the bindings holding the swarm-bot structure together would simply be released and the whole group would disaggregate and eventually reform into a different shape when a new mission is reassigned.
The aim of this paper is twofold: first introducing our project and second discussing how to apply the key characteristics of our swarm-bot, i.e., , team work, limited global knowledge, and emerging common goal, in order to achieve a service. The work is organized in the following sections: an introduction of the research context within which our research fits in (section 2), a brief introduction of our project (section 3), a more detailed description of our s-bots (section 4), a brief presentation of the 3D swarm-bot simulator developed (section 5), and a discussion of how a swarm-bot might be employed in order to carry out a specific service (section 6). Conclusions are drawn in section 7.
2 RELATED WORK
Considering the characteristics of a swarm-bot outlined above in the introduction, we could identify three research areas into which locating our concept: selfreconfiguration, self-assembly, and robot mobility.
In the first area, i.e., that of physical selfreconfiguration, researchers have put great emphasis essentially on the task of dynamical reshaping of physical structures by means of simple units [3, 4, 5].
Because the research focus in this topic has mainly been the reconfiguration itself, the aforementioned
simple units are not really entities independent from each other. As exemplified for instance in [6, 7, 8],
these units could start their evolution to a goal structure only from a physically pre-assembled configuration.
An implication, this, which implicitly pre-sets the number of structure components. A swarm-bot is, in
this respect, more powerful, since it neither assumes to start its evolution toward a goal from an already
pre-built configuration nor to have a pre-set limit of components.
As far as self-assembly is concerned, research in this area has basically just concentrated on the distributed
algorithms needed for assuming certain loose formation ([9]), or maintaining a certain loose planar geometrical shape ([10, 11]). Swarm-bots share with such a line of research the idea of exploiting local sensing in order to achieve the overall control of the group. However, the bindings among the different units within a group are in general rather loose as compared with those encountered in a swarm-bot. S-bots might, in fact, establish physical connections with each other in order to reach a target configuration. Such a characteristic allows them to extend formation geometries in theory also into the third dimension.
Concerning robot mobility, there is a great deal of research being pursued stretching from mechanics to autonomous control. Our involment with mobility, and hence autonomy, stems from the fact that our s-bot units need to gather in order to assume a certain shape. This means that they need to be capable of moving about independently when they are not joined together into a structure. The solution chosen for our s-bots has in a sense a conceptual similarity with that implemented for the wheeled JPL mars rover ([12]). The difference, however, lies in the fact that our s-bots’ locomotion subsystem is fixed to the main body, whereas that of the aforementioned rover has a variable geometry.
3 SWARM-BOT PROJECT
Having briefly introduced the context within which our research fits in, let us now present our project: the Swarm-Bot1. It is a three years pan-european research collaboration, currently in its first year, co-funded both by the Commission of the European Communities and by the Swiss National Science Foundation. Its main objective is to study a novel approach to the hardware design, implementation, and use of self-assembling, selforganising, and metamorphic robotic systems called swarm-bots. A large part of the research has so far consisted partly of feasibility studies and partly of physically implementing an initial design.
Since it is clear that hardware and control policies development go hand in hand, a 3D dynamic simulator has been under development since the beginning with the intent of using it both as a testing benchmark for the hardware design and as a tool for creating new group control algorithms. In this respect, although the project is currently still at its infancy, the particular phase we are in, here at IDSIA, is the completion of a first 3D simulator prototype which can already be used for hardware design testing and for developing distributed control policies with our s-bots. Notice that parallel phases of physical hardware construction and of definition of the controlling algorithms are also currently being carried out by the project partners at the Autonomous Systems Laboratory (LAS) of the EPFL in
