Distributed Planning and Control for Applications in Transportation Scheduling

Abstract

Abstract : The research addresses combinatorial problems in transportation planning and scheduling. This work addresses fundamental problems in real time planning and crisis decision making. Problems of uncertainty and incomplete information are dealt with by employing a stochastic domain model called a Markov decision process (MDP). Complexity in decision making is dealt with by using iterative techniques and decision theory to allocate computational resources at runtime. MASE is a simulation and development environment for problems involving multiple interacting agents. MASE provides robust and efficient communication and negotiation facilities to support distributed solutions to planning and utilities problems. MASE or its underlying communication subsystem running in a stand alone configuration can be used by distributed components implemented in C, C++, or Common Lisp. Discusses a complementary approach to envelope based methods for solving large MDPs, i.e., MDPs involving a very large number of states. This approach works by decomposing a large MDP into several smaller MDPs which are weakly coupled. A solution is obtained by solving each of the smaller MDPs individually and then combining these local solutions to obtain a global solution. An overview is provided on a proposed new direction of research for embedded planning and scheduling. Described is a general model for embedded planning and control systems. This model describes the rudiments of a software specification framework that is critical to any significant progress on real time systems of any complexity.