David J. Montana

Strongly typed genetic programming

Genetic programming is a powerful method for automatically generating computer programs via the process of natural selection (Koza, 1992). However, in its standard form, there is no way to restrict the programs it generates to those where the functions operate on appropriate data types. In the case when the programs manipulate multiple data types and contain functions designed to operate on particular data types, this can lead to unnecessarily large search times and/or unnecessarily poor generalization performance. Strongly typed genetic programming (STGP) is an enhanced version of genetic programming that enforces data-type constraints and whose use of generic functions and generic data types makes it more powerful than other approaches to type-constraint enforcement. After describing its operation, we illustrate its use on problems in two domains, matrix/vector manipulation and list manipulation, which require its generality. The examples are (1) the multidimensional least-squares regression problem, (2) the multidimensional Kalman filter, (3) the list manipulation function NTH, and (4) the list manipulation function MAPCAR.

Contact stability for two-fingered grasps

Two types of grasp stability, spatial grasp stability and contact grasp stability, each with a different concept of the state of a grasp, are distinguished and characterized. Examples are presented to show that spatial stability cannot capture certain intuitive concepts of grasp stability and hence that any full understanding of grasp stability must include contact stability. A model of how the positions of the points of contact evolve in time on the surface of a grasped object in the absence of any external force or active feedback is then derived. From the model, a general measure of the contact stability of any two-fingered grasp is obtained. Finally, the consequences of this stability measure and a related measure of contact manipulability on strategies for grasp selection are discussed. >

The condition for contact grasp stability

The author distinguishes between two types of grasp stability, called spatial grasp stability and contact grasp stability. The former is the tendency of the grasped object to return to an equilibrium location in space; the latter is the tendency of the points of contact to return to an equilibrium position on the objects surface. It is shown, via examples, that spatial stability cannot capture certain intuitive concepts of grasp stability, and hence that any full understanding of grasp stability must include contact stability. A model of how the positions of the points of contact evolve in time on the surface of the grasped object in the absence of any external force or active feedback is derived. From this model, a condition is obtained which determines whether or not a two-fingered grasp is contact stable.<<ETX>>

Genetic algorithms for complex, real-time scheduling

Real-time scheduling of large-scale problems in complex domains presents a number of difficulties for search and optimization techniques, including: large and complex search spaces; dynamically changing problems; and a variety of problem-dependent constraints and preferences. Genetic algorithms are well suited to such problems due to their adaptability and their effectiveness at searching large spaces. We have used genetic algorithms to solve real-world problems in areas such as field service scheduling, air crew scheduling and transportation scheduling. We discuss key aspects of our approach including: domain-specific chromosome representation and genetic operators; multi-objective evaluation function; heuristic initialization of the population; dynamic rescheduling; and cooperative interaction with human operators.

Optimizing parameters of a mobile ad hoc network protocol with a genetic algorithm

Mobile ad hoc networks are typically designed and evaluated in generic simulation environments. However the real conditions in which these networks are deployed can be quite different in terms of RF attentution, topology, and traffic load. Furthermore, specific situations often have a need for a network that is optimized along certain characteristics such as delay, energy or overhead. In response to the variety of conditions and requirements, ad hoc networking protocols are often designed with many modifiable parameters. However, there is currently no methodical way for choosing values for the parameters other than intuition and broad experience. In this paper we investigate the use of genetic algorithms for automated selection of parameters in an ad hoc networking system. We provide experimental results demonstrating that the genetic algorithm can optimize for different classes of operating conditions. We also compare our genetic algorithm optimization against hand-tuning in a complex, realistic scenario and show how the genetic algorithm provides better performance.

The kinematics of contact with compliance

The kinematics of contact describes the motion of a point of contact over the surfaces of two contacting objects in response to a relative motion of these objects. In a previous work (Int. of Robotics Res., vol.3, p.17-32, 1988), the author derived equations that embody this relationship when the two objects are assumed to be rigid bodies. In the present work, he extends that analysis by dropping the assumption of rigidity. He derives a set of equations, called the compliant contact equations, which model the kinematics of contact with compliance. He discusses an example that illustrates the effect of compliance on the kinematics of contact. He measures the trajectory of the center of contact on a tactile sensor in response to a known motion and shows how the results fit the proposed model. He analyzes how two tasks that are based on the rigid-body model of the kinematics of contact have been designed to be robust with respect to compliance. For the task of contour-following he provides experimental results that confirm this analysis.<<ETX>>

Coordination and control of multiple autonomous vehicles

The DARPA SIMNET project allows hundreds of soldiers to train together in a virtual air, land, and sea environment through a network of interactive simulators. In addition to the manned simulators, the virtual environment is also populated by a large number of autonomous vehicles called semi-automated forces, which are controlled by an operator at a single workstation. The authors address the issues of collision avoidance and formation keeping. The autonomous vehicles are responsible for the lower-level path planning, collision avoidance, and formation following. Routines are described for maneuvering among large obstacles, smaller objects, and moving vehicles.<<ETX>>

Evolution-Based Deliberative Planning for Cooperating Unmanned Ground Vehicles in a Dynamic Environment

Many challenges remain in the development of tactical planning systems that will enable automated, cooperative replanning of routes and mission assignments for multiple unmanned ground vehicles (UGVs) under changing environmental and tactical conditions. We have developed such a planning system that uses an evolutionary algorithm to assign waypoints and mission goals to multiple UGVs so that they jointly achieve a set of mission goals. Our evolutionary system applies domain-specific genetic operators, termed tactical advocates because they capture specific tactical behaviors, to make targeted improvements to plans. The plans are evaluated using a set of tactical critics that together comprise a multiobjective fitness function. Each critic evaluates a plan against criteria such as avoiding an enemy or meeting mission goals. Experimental results show that this approach produces highquality plans with the potential for real-time dynamic replanning.

Introduction to the special issue: Evolutionary algorithms for scheduling

Scheduling is a problem that occurs in a large variety of forms with huge cumulative economic and social consequences. Proper scheduling can provide better utilization of scarce and expensive resources as well as higher satisfaction for individuals such as customers and employees. For this reason, there has been a great deal of effort expended over many years on algorithms for automatic scheduling. However, except for a few isolated successes on some specialized problems, automatic scheduling remains an open problem. There are a few reasons why scheduling is such a difficult problem. One is the size and complexity of the search space. For the generic problem of assigning N tasks to M resources with a particular ordering of tasks at each resource, the number of possible solutions is

A multiagent society for military transportation scheduling

We are in the process of building a proof-of-concept automated system for scheduling all the transportation for the United States military down to a low level of detail. This is a huge problem currently handled by many hundreds of people across a large number and variety of organizations. Our approach is to use a multiagent society, with each agent performing a particular role for a particular organization. Use of a common multiagent infrastructure allows easy communication between agents, both within the transportation society and with external agents generating transportation requirements. We have demonstrated the feasi-bility of this approach on several large-scale deployment scenarios. Copyright