Swarmanoid communication

Introduction

The design of the communication system for the swarmanoid is a complex enterprise. Communication is important to regulate the activities of possibly heterogeneous groups, and to achieve the coordination of the swarmanoid as a whole.

When designing the communication system, some important aspects need to be taken into account:

• the expected role of communication: we identified three possible roles that communication can have within the swarmanoid: (i) information exchange, (ii) behaviour manipulation and (iii) entrainment, i.e., continuous bidirectional interactions. Identifying the role of communication in the achievement of a given goal is particularly important to devise the correct control structures and communication modalities. For example, in case the role of communication is identified as behaviour manipulation (e.g., an eye-bot that instructs the foot-bots on the position of an item to retrieve), a monodirectional communication channel is sufficient and should be preferred over a bi-directional channel. On the contrary, in order to synchronise the activities of a group, a bi-directional channel is required.
• the design methodology: The design methodology as well has a bearing on the selection of the correct communication modaliyty. We identified mainly two possible methodologies: (i) behaviour based techniques and (ii) evolutionary learning techniques. When a behaviour based technique is to be used, explicit and highly informative signals are generally preferred. For example, signals that encode a preferential direction can be efficiently exploited by a human designer that can directly decipher the information encoded into the signal. On the contrary, when artificial evolution is in charge of shaping the communication, simple global signals can be efficiently exploited.
• the hardware constraints: Hardware constraints constitute another important aspect to be taken into account. Depending on the task to be performed and on the expected role of communication, some device may be more appropriate than others. However, the choice of the communication device may constrain some of the dimensions of the communication, and may also require some particular control mechanism.

The above asepcts influence the possible dimensions on which the communication channel can vary. We propose four dimensions that encompass different communication features:

1. cardinality: the number of individuals addressed by an emitted signal (e.g., one-to-one vs. one-to-all communication).
2. expressivity: the amount of information that can be reliably transmitted over the communication channel.
3. spatiality: the range and directionality of the communication signals (e.g., local vs global communication).
4. modality: the quality of the interaction between the communicating agents (e.g., implicit vs. explicit, direct vs. indirect communication).

In conclusion, when designing the communication system, the important aspects to take into account are (i) the expected role of communication, (ii) the design methodology and (iii) the hardware constraints. All these aspects have a bearing on the four dimensions of communication discussed above, so that different solutions may be explored within the degrees of freedom not constrained by the above experimental choices.

Swarm intelligent design of the communication system

Taking inspiration from biological systems, such as ant colonies or honey bee swarms, robust and efficient control and communication system can be developed. This approach is characterised by an initial analytical phase in which models of the phenomena observed in nature are developed, in order to find out which are the basic mechanisms and individual interactions that govern a certain group behaviour. This knowledge is then exploited in the design phase, in which the discovered mechanisms are encoded into the control and communication systems. For example, a much studied form of communication in ants consists in laying pheromone trails that can be exploited to find the shortest path between the nest and a foraging area. A similar mechanism can be used by a robotic system, in which robots form a chain from the initial to the goal location.

The swarm intelligent approach to the study of the swarmanoid communication system will take inspiration from mass communication mechanisms observed in social insects (e.g., pheromone trails in ants) as well as forms of explicit and direct communication (e.g., waggle dance in honey bees, tandem runs in ants). Additionally, forms of communication that are not directly observed in biological systems will be also investigated whenever they can support the development of swarm intelligent control systems. In this respect, we consider all mechanisms that allow indicating a preferential direction or mechanisms for pointing.

Normally, within this approach to the study of communication, we focus on the development of the required communication protocols that can lead to certain group dynamics. For this reason, the communication system is normally hand-crafted and developed together with the control system.

Back to the introduction

Evolution of communication in embodied agents

Communication and control are strongly intertwined, and studying them in isolation may result in less efficient solutions. The evolutionary approach to the study of communication aims at avoiding any a priori decomposition of the control and communication aspects. It rather aims at contextually developing the behavioural and communication strategies, which can co-evolve as a single whole. This opens the way to complex forms of communication that are tightly linked with the sensory-motor coordination of the robots and their cognitive abilities (e.g., integration over time of perceptual information for decision making). In some cases, the evolutionary approach produces communication protocols that proved more adapted than hand-crafted ones. This is possible because the communicative and non communicative behaviours co-evolve and adapt one to the other, exploiting the fine-grained interactions between the robots and the physical and social environment.

Within this approach, we aim at studying the evolution of the behavioural and cognitive features that underpin complex forms of communication. Moreover, we are interested in studying the selective advantage that communication can have in a group of possibly heterogeneous robots. Additionally, we plan to investigate the evolution of simple forms of communication (e.g., indirect and/or implicit communication) that can support selforganising processes and collective decision making.

As a design methodology, we mainly focus on genetic evolution. Social learning and cultural evolution are alternative techniques that may be considered within the Swarmanoid project. However, the possibility to apply also these techniques will be evaluated later on in the course of the project.

Back to the introduction

The networked swarm

All the different robots composing the swarmanoid have a set of radio wireless interfaces. The combined use of these interfaces can allow to transmit across the swarm relatively large volumes of structured information which can be used for the effective coordination of the activities of the swarm itself. In the activities of the workpackage Communication, we aim at studying how structured information sent over the radio channels can be reliably and effectively transmitted and used to coordinate the fully distributed generation of systemlevel behaviours. In particular, we are interested in studying the use of: (i) networking techniques to represent and spread useful information within the robot network, (ii) networking and connectivity information itself to direct robot decisions and global coordination in relationship to basic issues such as navigation toward a target and automatic task division. Generally speaking, setting and use of a radio network in a multi-robot system pertain to the rather novel domain of so-called networked robotics. Mobile networked robots such as the swarmanoid, pose a number of technical challenges related to network noise, reliability, congestion, topological and delay variability, link stability, range and power limitations, deployment, coverage, safety, localisation, sensor and actuation fusion at robotand system-level, etc.. In the Swarmanoid project we specifically address only a limited subset of the these issues in the perspective of providing solutions which are in the spirit of the swarm intelligence view. That is, fully distributed self-organising strategies which do not rely on ad hoc architectural or hardware assumptions, are highly scalable and adaptive with respect to the number and the characteristics of the robots, and can show robust behaviour to dynamic changes and failures.

Back to the introduction