Research and technology developments on autonomous agents and autonomic computing promote a vision of artificial systems that are able to resiliently manage themselves and autonomously deal with issues at runtime in dynamic environments. Indeed, autonomy can be leveraged to unburden humans from mundane tasks (cf. driving and autonomous vehicles), from the risk of operating in unknown or perilous environments (cf. rescue scenarios), or to support timely decision-making in complex settings (cf. data-centre operations). Beyond the results that individual autonomous agents can carry out, a further opportunity lies in the collaboration of multiple agents or robots. Emerging macro-paradigms provide an approach to programming whole collectives towards global goals. Aggregate computing is one such paradigm, formally grounded in a calculus of computational fields enabling functional composition of collective behaviours that could be proved, under certain technical conditions, to be self-stabilising. In this work, we address the concept of collective autonomy, i.e., the form of autonomy that applies at the level of a group of individuals. As a contribution, we define an agent control architecture for aggregate multi-agent systems, discuss how the aggregate computing framework relates to both individual and collective autonomy, and show how it can be used to program collective autonomous behaviour. We exemplify the concepts through a simulated case study, and outline a research roadmap towards reliable aggregate autonomy.

No abstract

Recent trends in pervasive computing promote the vision of Collective Adaptive Systems (CASs): large-scale collections of relatively simple agents that act and coordinate with no central orchestrator to support distributed applications. Engineering global behaviour out of local activity and interaction, however, is a difficult task, typically addressed by try-and-error approaches in simulation environments. In the context of Aggregate Computing (AC), a prominent functional programming approach for CASs based on field-based coordination, this difficulty is reflected in the design of versatile algorithms preserving efficiency in a variety of environments. To deal with this complexity, in this work we propose to apply Machine Learning techniques to automatically devise local actions to improve over manually-defined AC algorithms specifications. Most specifically, we adopt a Reinforcement Learning-based approach to let a collective learn local policies to improve over the standard gradient algorithm-a cornerstone brick of several higher-level self-organisation algorithms. Our evaluation shows that the learned policies can speed up the self-stabilisation of the gradient to external perturbations.

No abstract

Field-based coordination is a model for expressing the coordination logic of large-scale adaptive systems, composing functional blocks from a global perspective. As for any coordination model, a proper toolchain must be developed to support its adoption across all development phases. Under this point of view, the ScaFi toolkit provides a coordination language (field calculus) as a DSL internal in the Scala language, a library of reusable building blocks, and an infrastructure for simulation of distributed deployments. In this work, we enrich such a toolchain by introducing ScaFi-Web, a web-based application allowing in-browser editing, execution, and visualisation of ScaFi programs. ScaFi-Web facilitates access to the ScaFi coordination technology by flattening the learning curve and simplifying configuration and requirements, thus promoting agile prototyping of field-based coordination specifications. In turn, this opens the door to easier demonstrations and experimentation, and also constitutes a stepping stone towards monitoring and control of simulated/deployed systems.