Control >> Functional self-assembly

Functional self-assembly

One of the most interesting features of the s-bots is their capability to self-assemble in a bigger structure (i.e., a swarm-bot) which can carry out tasks that can hardly be performed by a single s-bot. Self-assembling is made possible by the morphological structure of the s-bots; that is, (i) owing to the fixed and flexible arms the robots can grasp one to each other; (ii) owing to a variety of sensors the agents can coordinate their actions in order to cooperatively carry out a task. The reader has probably already noticed that the issue of self-assembling is not entirely circumscribed to the hardware design of the robots. That is, the design of control structures for self-assembling robots represents a challenging exercise. For example, the connection among s-bots that move one respect to the other requires such a coordination of actions which, from the experimenter point of view, are difficult to design. Moreover, self-assembling does not only concerns the emergence of those coordinated movements that make two or more s-bots connect to each other. In fact, decision making mechanisms are involved as well.  In particular, any single s-bot should be equipped with the mechanisms required to figure out when individual actions are less efficient than collective ones. In the context of the SWARM-BOTS project we have extensively studied self-assembling by focusing on the dynamics of the coordinated actions required to the s-bots to connect to each other, by excluding the decision making mechanisms above mentioned. In particular, self-assembling has been studied in the context of prey retrieval. In this kind of scenario, the s-bots are placed close to a prey that is too heavy to be moved by a single robot. Therefore, in order to retrieve the prey, a group of s-bots is required to form multiple connections to the prey and, in those cases in which the perimeter of the prey is not big enough to enable all the s-bots to grasp it, they should also be able to connect to each other. No decision is required in order to figure out whether the individual actions could be as efficient as the collective ones. The focus is on how the s-bots manage to coordinate their actions in order to connect to each other and to the prey to move it as far as possible (see ">here).


Here, we present preliminary work on functional self-assembling, that is, a scenario in which self-assembling requires decision making mechanisms. Functional self-assembling refers to the self-organised creation of a physically connected structure, which should be functional to the accomplishment of a particular task.  In other words, our goal is the design of controllers for s-bots capable of connecting to each other (i.e., forming a swarm-bot) any time environmental contingencies prevent the single s-bot to achieve its goal.

 

Figure 1. A graphical representation of the task.

The task requires navigation within a rectangular corridor in order to approach light bulbs positioned on the opposite end with respect to the s-bots' starting positions (see Figure 1). The corridor (4 meters long, 1 meter wide) is divided in an area of high temperature and an area of low temperature (respectively, light and dark gray in Figure 1). Aggregation and assembling are required in order to traverse a low temperature area, within which a swarm-bot (i.e., assembled s-bots) navigates more effectively than a group of disconnected s-bots. The effectiveness of the navigational strategies is correlated with the amount of "energy" required by the s-bot to explore the corridor. In the area of high temperature, each s-bot saves more of its energy by navigating disconnected, while in the area of low temperature, each s-bot saves more energy by navigating assembled---i.e., by forming a swarm-bot. If, while navigating, an s-bot exhausts its energy, it is not able to move any more. The s-bots do not have any information concerning their energy level.  However, the s-bots can reach the light bulbs before running out of energy if they properly react to the characteristics of the environment. In particular, an optimal strategy requires the s-bots (i) to individually move toward the light bulbs as long as the temperature remains high; (ii) to aggregate by exploiting the sound signalling system they are provided with as soon as the temperature drops; (iii) to continue their phototactic behaviour in an assembled structure (i.e., by forming a swarm-bot) throughout the low temperature area.

Homogeneous groups of s-bots are controlled by artificial neural networks, whose parameters are set by an evolutionary algorithm. A single genotype is used to create a group of individuals with an identical control structure. Each s-bot is controlled by a fully connected, 14 neuron continuous time recurrent neural network.

Results

Example movies:
• performing functional self-assembling (MOV 0.8MB).

• V. Trianni, E. Tuci, and M. Dorigo. Evolving functional self-assembling in a swarm of autonomous robots. In S. Schaal, A. Ijspeert, A. Billard, S. Vijayakamur, J. Hallam, and J.-A. Meyer, editors, From Animals to Animats 8. Proceedings of the Eight International Conference on Simulation of Adaptive Behavior (SAB04), pages 405-414. MIT Press, Cambridge, MA, 2004.

Control >> Functional self-assembly