The eight research papers in this volume are being published forthe third time. They were originally presented as a special session onrobotic agents at the First International Conference on AutonomousAgents (Agents ’97), and short versions appeared in the proceedings ofthat conference. Bekey, the coordinator of that session, later edited aspecial issue of Kluwer’s journal AutonomousRobots containing full versions of the papers. Thisvolume is a reprint of that special issue. It contains no index,cumulative bibliography, or other material beyond what one would expectin an issue of a research journal. All eight papers describetechnologies that can enable robots to operate to some degreeindependently of their masters.

Fujita and Kitano, in “Development of an Autonomous QuadrupedRobot for Robot Entertainment,” discuss MUTANT, a dog-like robotwith a layered architecture that integrates instincts and emotions,high-level cognition, and reactive behavior. MUTANT interacts withpeople both visually and verbally, using a tone-based system forunderstanding spoken commands, and supports behaviors such as tracking aball, shaking hands, and sleeping.

Garcia-Alegre and Recio, in “Basic Visual and Motor Agentsfor Increasingly Complex Behavior Generation on a Mobile Robot,”present a set of behavioral modules for locomotion (such asSTOP&BACKWARD, AVOID, and FORWARD) and vision (such as SACCADIC,FIND_CONTOUR, and CENTER). High-level modules (forexample, following a person or leaving a room) are programmed in termsof lower-level ones. The use of the term “agent” for suchlow-level modules seems forced, especially in view of their lack ofautonomy.

“An Autonomous Spacecraft Agent Prototype,” by Pell etal, describes the New Millennium Remote Agent architecture forautonomous spacecraft control. This hybrid multithreaded architecturecombines real-time monitoring and control with components that supportconstraint-based planning and scheduling and model-based diagnosis andreconfiguration. The ability of the architecture to combine deliberationwith real-time responsiveness is demonstrated in a scenario in which aspacecraft must manage its own insertion into orbit around Saturn.

Lopez-Sanchez et al, in “Map Generation by CooperativeLow-Cost Robots in Structured Unknown Environments,” demonstratethe use of a swarm of expendable independent robots to map orthogonalwalls in an unknown environment. Each robot wanders around theenvironment, builds a partial map based on its own observations, andexchanges partial maps with any other robot it encounters. The home baseintegrates partial maps from those robots that make it back home, usingfuzzy logic to decide whether two reported segments represent the samewall or different walls.

Horswill, in “Grounding Mundane Inference inPerception,” addresses the problem of coupling reactive anddeliberative systems without falling off the cliff of computationalcomplexi ty. He exhibits a parallelizable subset of modal logic whosevariables are indexed based on linguistic case grammar, and shows how itcan be used to integrate perception and routine reasoning for a simpledelivery robot.

Haigh and Veloso’s paper “Interleaving Planning and RobotExecution for Asynchronous User Requests” describes ROGUE, anarchitecture that integrates the PRODIGY4.0 domain-independent plannerand the Xavier robot by interleaving planning and execution in a waythat supports asynchronous user requests.

Durfee, Kenny, and Kluge, in “Integrated Permission Planningand Execution for Unmanned Ground Vehicles,” seek to enhance theoperator interface for unmanned ground vehicles (UGVs) for militaryapplications by permitting operators to task UGVs in military termsrather than using robotics concepts; increasing vehicle autonomy tominimize operator intervention; and supporting operation whencommunications are sporadic. They treat each legacy UGV controller andthe legacy operator interface as agents, whose interactions are guidedby a procedural encoding of domain knowledge.

“Learning View Graphs for Robot Navigation,” by Franzet al, shows how a robot can navigate in a previously unknownenvironment on the basis of a series of connected views. The adjacencyrelations between successive views provide a map that the robot can useto determine its position in terms of where it has been. Such mechanismsappear to underlie the ability of bees and ants to pinpoint a locationdefined by visual cues. Previous experiments combine visual informationwith metric data from compasses or wheel encoders. This paper exploresthe potential of using vision information alone.

Robotics has long been a helpful foil to more abstract forms ofagent research, because it forces agents to interact withreal-world constraints that are easily overlooked in simulatedenvironments. The papers in this collection offer a rich array oftechniques for balancing real-time operations such as sensing andreacting with computationally costly reasoning activities. However, mostof the examples in this volume concern only single robots, and thus donot address the challenges and opportunities presented by systems inwhich agents interact with one another.