Distributed Artificial Intelligence (DAI) has existed as a subfield of AI for less than two decades. DAI is

The goal of creating an integrated cognitive robot is still only a tantalizing dream. Current artificial intelligence and robotics research is highly divergent with little or no commonality among specialized subfields. New rich task domains are needed to pose the right challenges to extant theories and promote convergence. We propose soccerplaying as such a task since it requires situated robotics, perception, real-time decision making, planning, plan recognition, learning and multirobot coordination and control. The technology to perform real-time vision and build autonomous robots is available; the Dynamite testbed has been built to perform experiments with multiple robots.

Dynamic environments pose multiple problems for autonomous agents. Traditionally, agents were designed to develop an internal model of the world and use it to create plans to achieve their goals. Such a strategy fails when the actual environment differs from the stored representation. This realization lead to reactive agents, which stored no model of the world. Today, most researchers would agree that representation is not intrinsically bad, but the cost of keeping an arbitrary amount of representation up-to-date in a dynamic environment outweighs its benefits. We present a simple task-dependent representation system, called the effective field of view, which can be used to augment a reactive agent and improve its competence at a variety of tasks. We have embodied this system in two situated autonomous agents. With the first, we quantify the amount of representation necessary for its tasks, as well as the performance gains. With the second, we show how the effective field of view can be used in a hierarchical layered agent architecture. We show that the effective field of view can dramatically increase agent performance. 1.

Our mobile robot, Spinoza, embodies a sophisticated real-time vision system for control of a mobile robot in a dynamic environment. The complexity of our robot architecture arises from the wide variety of tasks that need to be performed and the resulting chal-lenge of coordinating multiple distributed, concurrent processes on a diverse range of processor architec-tures including Transputers, digital signal processors, and a workstation host. The system handles sens-ing, reasoning, and action components of a robot dis-tributed over these architectures, and responds to un-predictable events in an unknown dynamic environ-ment. Spinoza relies heavily on its capability to per-form real-time vision processing in order to perform task such as mapping, navigation, exploration, track-ing, and simple manipulation. 1

Robotic Soccer involves multiple agents that need to collaborate in an adversarial environment to achieve specific objectives. We have been building an architecture that addresses this integration of high-level and low-level reasoning as a combined system of mini-robots, a camera for perception, a centralized interface computer, and several client servers as the minds of the mini-robot players. In this paper we focus on the hardware design of our mini-robots. Our main purpose is to provide a detailed description of our design decisions so that others may learn from and replicate our efforts. Communication between the interface computer and the mini-robots is achieved through coded infrared radiation. The minirobots can turn on the spot and move forward and backward with variable speed. Our design also allows a well-balanced use of the on-board power and current supplies. 1 Introduction As robots become more adept at operating in the real world, the high-level issues of collaborative ...

Existing approaches to high-level robot control which support deliberation in one form or another usually do not view the time spent on reasoning as critical, which is true in typical applications like oce delivery. This is not so, however, in domains like robotic soccer where a team of robots must cooperate in a highly dynamic environment and where actions need to be chosen under tight resource constraints. For this reason, most existing systems for such domains rely on purely reactive architectures without a reasoning component. Our aim is to build robotic soccer agents which are capable of limited forms of deliberation using the action language Golog. In order to meet the real-time constraints, we propose a hybrid architecture which enables the robot to react to the environment very fast as well as to choose actions proposed by the reasoning component. The hybrid nature of our approach manifests itself not only in the dierence between high-level control and low-level routines such as self-localization, but also within the action selection process, that is, within the high-level controller itself.

Mobile robots must cope with uncertainty from many sources along the path from interpreting raw sensor inputs to behavior selection to execution of the resulting primitive actions. This article identifies several such sources and introduces methods for i) reducing uncertainty and ii) making decisions in the face of uncertainty. We present a complete vision-based robotic system that includes several algorithms for learning models that are useful and necessary for planning, and then place particular emphasis on the planning and decision-making capabilities of the robot. Specifically, we present models for autonomous color calibration, autonomous sensor and actuator modeling, and an adaptation of particle filtering for improved localization on legged robots. These contributions enable effective planning under uncertainty for robots engaged in goal-oriented behavior within a dynamic, collaborative and adversarial environment. Each of our algorithms is fully implemented and tested on a commercial off-the-shelf vision-based quadruped robot.

This work studies the control of robots in the adversarial world of "Hunt the Wumpus". The hybrid learning algorithm which controls the robots behavior is a combination of a modified RPNI algorithm, and a utility update algorithm. The modified RPNI algorithm is a DFA learning algorithm, to learn opponents' strategies. An utility update algorithm is used to quickly derive a successful conclusion to the mission of the agent using information gleaned from the modified RPNI. 1 Introduction Developing single purpose learning algorithms for manipulating robots in a given domain, has given way to using hybrid learning algorithms that yield a more robust behavior. An approach to combine a planner with a reactive agent to play the game of soccer has been explored in the literature (Sahota 1994). In this paper, we are interested in developing a hybrid learning algorithm to manipulate robots in an adversarial game, often known as "Hunt the Wumpus". In the wumpus world, there are two a...

Industrially deployed robots are currently very dependent on human guidance and support. They rely on operator intervention to maintain a stable, known environment within which they perform specific well-defined tasks. Such constraints limit the extent to which robots can be employed. Many useful applications await the development of self-reliant, autonomous robotsapplications ranging from exploring Mars to cleaning factory floors. New design techniques are evolving to meet this challenge. One technique that has met with considerable success is "behaviour-based" control. In this form of control, the modules or "behaviours" within the robot's control system take their cues primarily from the environment, rather than from each other as in more traditional schemes. Robots constructed in this way need to make fewer assumptions about their environment, and so are more robust. However, it can be difficult to control the interaction of such behaviours. In this thesis, a novel scheme for orchestrating the actions of collections of behaviours is presented, and implemented on a physical robot.

Robotic applications impose hard real-time demands on their vision components. To accommodate the real-time constraints, the visual component of robotic systems are often simplified by narrowing the scope of the vision system for a particular task. Another option is to build a generalized vision (sensor) processor and provides multiple interfaces, of differing scales and content, to other modules in the robot. Both options can be implemented in many ways, depending on computational resources. The tradeoffs among these alternatives become clear when we study the vision process as a server whose clients request information about the world. We model the interface on client-server relations in user interfaces and operating systems. We examine the relation of this model to robot and vision sensor architecture and explore its application to a variety of vision sensor implementations. 1 1 This research was supported by the Natural Sciences and Engineering Research Council of Canada and the ...