The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies or endorsements, either expressed or implied, of Carnegie Mellon University. The problem of efficient multirobot coordination has risen to the forefront of robotics research in recent years. Interest in this problem is motivated by the wide range of application domains demanding multirobot solutions. In general, multirobot coordination strategies assume either a centralized approach, where a single robot/agent plans for the group, or a distributed approach, where each robot is responsible for its own planning. Inherent to many centralized approaches are difficulties such as intractable solutions for large groups, sluggish response to changes in the local environment, heavy communication requirements, and brittle systems with single points of failure. The key advantage of centralized approaches is that they can produce globally optimal plans. While most distributed approaches can overcome the obstacles inherent to centralized approaches, they can only produce suboptimal plans. This work explores the development of a market-based architecture that will be inherently distributed, but will also opportunistically form centralized sub-groups to improve efficiency, and thus

The tcc model is a formalism for reactive concurrent constraint programming. We present a model of temporal concurrent constraint programming which adds to tcc the capability of modeling asynchronous and non-deterministic timed behavior. We call this tcc extension the ntcc calculus. We also give a denotational semantics for the strongest-postcondition of ntcc processes and, based on this semantics, we develop a proof system for linear-temporal properties of these processes. The expressiveness of ntcc is illustrated by modeling cells, timed systems such as RCX controllers, multi-agent systems such as the Predator /Prey game, and musical applications such as generation of rhythms patterns and controlled improvisation. 1

Traffic congestion is one of the leading causes of lost productivity and decreased standard of living in urban settings. Recent advances in artificial intelligence suggest vehicle navigation by autonomous agents will be possible in the near future. In this paper, we propose a reservationbased system for alleviating traffic congestion, specifically at intersections, and under the assumption that the cars are controlled by agents. First, we describe a custom simulator that we have created to measure the different delays associated with conducting traffic through an intersection. Second, we specify a precise metric for evaluating the quality of traffic control at an intersection. Using this simulator and this metric, we show that our reservation-based system can perform two to three hundred times better than traffic lights. As a result, it can smoothly handle much heavier traffic conditions. We show that our system very closely approximates an overpass, which is the optimal solution for the problem with which we are dealing.

This paper introduces AntNet, a new routing algorithm for telecommunication networks. AntNet is an adaptive, distributed, mobile-agents-based algorithm which was inspired byrecent work on the ant colony metaphor. We apply AntNet in a datagram network and compare it with both static and adaptive state-ofthe-art routing algorithms. We ran experiments for various paradigmatic temporal and spatial tra c distributions. AntNet showed both very good performances and robustness under all the experimental conditions with respect to its competitors. 1

Abstract not found

We present an adaptation of the standard genetic program (GP) to hierarchically decomposable, multi-agent learning problems. To break down a problem that requires cooperation of multiple agents, we use the team objective function to derive a simpler, intermediate objective function for pairs of cooperating agents. We apply GP to optimize first for the intermediate, then for the team objective function, using the final population from the earlier GP as the initial seed population for the next. This layered learning approach facilitates the discovery of primitive behaviors that can be reused and adapted towards complex objectives based on a shared team goal.

This thesis describes the incremental development and main features of a synthetic multi-agent system called UvA Trilearn 2001. UvA Trilearn 2001 is a robotic soccer simulation team that consists of eleven autonomous software agents. It operates in a physical soccer simulation system called soccer server which enables teams of autonomous software agents to play a game of soccer against each other. The soccer server provides a fully distributed and real-time multi-agent environment in which teammates have to cooperate to achieve their common goal of winning the game. The simulation models many real-world complexities such as noise in object movement, noisy sensors and actuators, limited physical abilities and restricted communication. This thesis addresses the various components that make up the UvA Trilearn 2001 robotic soccer simulation team and provides an insight into the way in which these components have been (incrementally) developed. Our main contributions include a multi-threaded three-layer agent architecture, a flexible agent-environment synchronization scheme, accurate methods for object localization and velocity estimation using particle filters, a layered skills hierarchy, a scoring policy for simulated soccer agents and an e#ective team strategy. Ultimately, the thesis can be regarded as a handbook for the development of a complete robotic soccer simulation team which also contains an introduction to robotic soccer in general as well as a survey of prior research in soccer simulation. As such it provides a solid framework which can serve as a basis for future research in the field of simulated robotic soccer. Throughout the project UvA Trilearn 2001 has participated in two international robotic soccer competitions: the team reached 5th place at the German ...

In this paper we present two versions of AntNet, a novel approach to adaptive learning of routing tables in wide area best-effort datagram networks. AntNet is a distributed multi-agent system inspired by the stigmergy model of communication observed in ant colonies. We report simulation results for AntNet on realistically sized networks using as performance measures throughput, packet delays and resources utilization. Our tests show that both instances of AntNet show superior performance with respect to the current Internet routing algorithm (OSPF), some improved old Internet routing algorithms (SPF and distributed adaptive Bellman-Ford), and recently proposed forms of asynchronous online Bellman-Ford (Q-routing and Predictive Q-routing). KEYWORDS: Adaptive routing, ant colony optimization, distributed multi-agent systems. 1 INTRODUCTION In this paper we consider the problem of adaptive routing in communications networks: we focus on routing for wide area datagram networks with irre...

Ant-like systems take advantage of agents ' situatedness to reduce or eliminate the need for centralized control or global knowledge. This reduces the need for complexity ofindividuals and leads to robust, scalable systems. Such insect-inspired situated approaches have proven e ective both for task performance and task allocation. The desire for general, principled techniques for situated interaction has led us to study the exploitation of abstract situatedness { situatedness in non-physical environments. The port-arbitrated behavior-based control approach provides a wellstructured abstract behavior space in which agents can participate in situated interaction. We focus on the problem of role assumption, distributed task allocation in which each agent selects its own task-performing role. This paper details our general, principled Broadcast of Local Eligibility (BLE) technique for role-assumption in such behavior-space-situated systems, and provides experimental results from the CMOMMT target-tracking task.

In this paper we focus on the problem of designing a collective of autonomous agents that individually learn sequences of actions such that the resultant sequence of joint actions achieves a predetermined global objective. We are particularly interested in instances of this problem where centralized control is either impossible or impractical. For single agent systems in similar domains, machine learning methods (e.g., reinforcement learners [18]) have been successfully used [1, 2, 3, 31]. However, applying such solutions directly to multi-agent systems often proves problematic, as agents may work at cross-purposes, or have difficulty in evaluating their contribution to achievement of the global objective, or both. Accordingly, the crucial design step in multiagent systems centers on determining the private objectives of each agent so that as the agents strive for those objectives, the system reaches a good global solution. In this work we consider a version of this problem involving multiple autonomous agents in a grid world. We use concepts from collective intelligence [19, 27, 30] to design goals for the agents that are "aligned" with the global goal, and are "learnable" in that agents can readily see how their behavior affects their utility. We show that reinforcement learning agents using those goals outperform both "natural" extensions of single agent algorithms and global reinforcement learning solutions based on "team games".