Multi-agent system

About: Multi-agent system is a research topic. Over the lifetime, 27978 publications have been published within this topic receiving 465191 citations. The topic is also known as: multi-agent systems & multiagent system.

The Stability, Scalability and Performance of Multi-agent Systems

Abstract: Much has been published on the functional properties of multi-agent systems (MASs) including their co-ordination rationality and knowledge modelling. However, an important research area which has so far received only scant attention covers the non-functional properties of MASs which include performance, scalability and stability issues — clearly thes become increasingly important as the MAS field matures, and as more practical MASs become operational. An understanding of how to evaluate and assess such non-functional properties, and hence how to improve on them by altering the underlying MAS design, is gradually emerging as a pressing concern. This paper presents preliminary work to address such concerns; particularly, it investigates the performance and scalability of a multi-agent model we have developed. Firstly, this paper defines performance, scalability and stability within the context of multi-agent applications. Following, we describe a multi-agent model that we later use to illustrate our first attempts at evolving a procedure for analysing such non-functional properties of MASs. Next, we report on our initial attempts to investigate the performance and scalability of the multi-agent model. Finally, the significance of these results in particular and of such investigations in general is discussed.

Synchronization and control of complex networks via contraction, adaptation and evolution

Abstract: Complex networked systems abound in Nature and Technology. They consist of a multitude of interacting agents communicating with each other over a web of complex interconnections. Flocks of birds, platoon of cooperating robots, swirling fishes in the Ocean are all examples whose intricate dynamics can be modeled in terms of three essential ingredients: (i) a mathematical description of the dynamical behavior of each of the agents in the network; (ii) an interaction (or coupling) protocol used by agents to communicate with each other and (iii) a graph describing the network of interconnections between neighboring agents. These three elements are actually mapped onto the mathematical model usually considered in the literature to describe a complex network which uses appropriate equations to describe the node dynamics, the coupling protocol and the network topology.

PRIMAL: Pathfinding via Reinforcement and Imitation Multi-Agent Learning

Abstract: Multi-agent path finding (MAPF) is an essential component of many large-scale, real-world robot deployments, from aerial swarms to warehouse automation. However, despite the community's continued efforts, most state-of-the-art MAPF planners still rely on centralized planning and scale poorly past a few hundred agents. Such planning approaches are maladapted to real-world deployments, where noise and uncertainty often require paths be recomputed online, which is impossible when planning times are in seconds to minutes. We present PRIMAL, a novel framework for MAPF that combines reinforcement and imitation learning to teach fully decentralized policies, where agents reactively plan paths online in a partially observable world while exhibiting implicit coordination. This framework extends our previous work on distributed learning of collaborative policies by introducing demonstrations of an expert MAPF planner during training, as well as careful reward shaping and environment sampling. Once learned, the resulting policy can be copied onto any number of agents and naturally scales to different team sizes and world dimensions. We present results on randomized worlds with up to 1024 agents and compare success rates against state-of-the-art MAPF planners. Finally, we experimentally validate the learned policies in a hybrid simulation of a factory mockup, involving both real world and simulated robots.

Distributed Output-Feedback Control of Nonlinear Multi-Agent Systems

Abstract: This technical note presents a cyclic-small-gain approach to distributed output-feedback control of nonlinear multi-agent systems. Through novel distributed observer and control law designs, the closed-loop multi-agent system is transformed into a large-scale system composed of input-to-output stable (IOS) subsystems, the IOS gains of which can be appropriately designed. By guaranteeing the IOS of the closed-loop multi-agent system with the recently developed cyclic-small-gain theorem, the outputs of the controlled agents can be driven to within an arbitrarily small neighborhood of the desired agreement value under bounded external disturbances. Moreover, if the system is disturbance-free, then asymptotic convergence can be achieved. Interestingly, the closed-loop distributed system is also robust to bounded time-delays of exchanged information.