Harold C. Forbes

Io, Ganymede, and Callisto A Multiagent Robot Trash-Collecting Team

The Georgia Institute of Technology won the Office Cleanup event at the 1994 AAAI Robot Competition and Exhibition with a multirobot cooperating team. This article describes the design and implementation of these reactive trash-collecting robots, including details of multiagent cooperation, color vision for the detection of perceptual object classes, temporal sequencing of behaviors for task completion, and a language for specifying motor schema-based robot behaviors.

Intelligent control in traffic management

The goal of an advanced traffic management system (ATMS) is to efficiently manage existing transportation resources in response to dynamic traffic conditions. The utility of an ATMS will greatly depend upon its ability to adaptively respond to traffic patterns and permutations. The application of knowledge-based systems and neural networks provides an ATMS with the technology required to control traffic in an intelligent manner. The volume of traffic combined with the number of streets and intersections an operator control station must monitor clearly dictates the need for computer support. Integrating these technologies with existing transportation methodologies produces a semi-autonomous system capable of reducing operator workloads while maintaining high levels of safety. This paper describes an intelligent traffic management control system called TERMINUS developed to adaptively respond to real-time traffic management problems.<<ETX>>

Object technologies and real-time scheduling

Any object-based real-time system gives rise to real-time scheduling issues not easily solved by use of standard scheduling methods, including: (1) the invocation of an objects method may imply the need for online schedulability analysis for the thread(s) able to execute method code, (2) parallelism internal to objects may result in the need to schedule multiple threads and target processors simultaneously, and (3) methods that invoke other methods as part of their execution give rise to hierarchically composed groups of to-be-scheduled real-time threads. The Rapid scheduler presented in this paper addresses such dynamic object-based real-time systems for target machines comprised of both multi- and uniprocessor machines. To address the variety of needs of object-based real-time systems, Rapid is 1) configurable to the target multiprocessor architecture, 2) dynamically re-configurable to changes in application requirements, 3) efficient in terms of processor utilization, and 4) effective in producing high quality schedules. Rapids current implementation addresses hard deadline application, with future work concerning alternative formulations of timing constraints.

Unmanned guided vehicle system

Unmanned guided vehicles (UGV) require the ability to visually understand the objects contained within their operating environments in order to locally guide vehicles along a globally determined route. Several large scale programs have been funded over the past decade that have created multimillion dollar prototype vehicles incapable of functioning outside of their initial test track environment. This paper describes the Unmanned Guided Vehicle System (UGVS) developed for the US Army Missile Command for operation in natural terrain. The goal of UGVS is to develop a real-time system adaptive to a range of terrain environments (e.g. roads, open fields, wooded clearings, forest areas) and seasonal conditions (e.g., fall, winter, summer, spring). UGVS consists of two primary processing activities. First, the UGVS vision system is tasked with determining the location of gravel roads in video imagery, detecting obstacles in the vehicles path, identifying distant road spurs, and assigning a classification confidence to each image component. Second, the guidance and navigation system computes the global route the vehicle should pursue, utilizes image classification results to determine obstructions in the local vehicle path, computes navigation commands to drive the vehicle around hazardous obstacles, correlates visual road spur cues with global route digital maps, and provides the navigation commands to move the vehicle forward. Results of UGVS working in a variety terrain environments are presented to reinforce system concepts.

Dynamic scheduling for mobile robots

This research concerns efficient multiprocessor threads-based implementation of reactive navigation for mobile robots. We present two important results: 1) Performance is improved significantly when CPU time allocated to individual navigational threads is adjusted dynamically according to a heuristic measure of their importance. 2) To implement this strategy, we present a multiprocessor scheduler design which can dynamically schedule navigational threads. The experiments were conducted in simulation on a BBN Butterfly and a KSR1 (shared memory multiprocessors). Speedups found for this example should extend to more complex navigational strategies as long as a heuristic measure of thread importance is available.<<ETX>>