Multi-Agent Planning for Advanced Traffic Management

This paper describes two multi-agent systems developed at Southwest Research Institute to address issues in automated planning. The Autonomous Underwater Vehicle (AUV) Planning system uses multiple cooperative agents to generate a path for the AUV. The Medical Test Planning System (MTPS) consists multiple agents that incrementally build a schedule for a series of medical tests constrained by the resources available to perform them. The approaches used by these systems have considerable applicability to the problems faced in the area of advanced traffic management. The Role of Planning and Scheduling in IVHS One of the primary issues in the Intelligent Vehicle Highway Systems 0VHS) program is traffic synchronization. This issue encompasses requirements for automated checkin to the Automated Highway System (AHS) well as requirements for lateral and longitudinal control of traffic flow. It also is related to incident detection and management. These requirements define a real-time planning and scheduling problem. At check-in, the automated system establishes the operating envelop of the vehicle and its driver and the desired destination. This information can be used by an automated system for planning routes and scheduling platoons of vehicles. Incident detection could provide additional data to the system, placing constraints on available routes and reasonable platoon schedules. The planning and scheduling problem associated with traffic synchronization in an intelligent vehicle highway system is extremely complex, consisting of multiple conflicting goals and constraints. The overall goal is to construct a plan that moves vehicles to their destination safely and efficiently. This implies that both highway safety and throughput will be increased when, in the past, they have essentially been opposing goals. Planning for IVHS entails the resolution of multiple constraints under circumstances of resource limitations. Southwest Research Institute (SwRI) has investigated similar planning problems and determined that one viable method of automating complex planning and scheduling problems is the use of multi-agent architectures. This paper presents two planning systems that utilize advanced software technology in support of the types of problems that will be encountered in automated traffic management. 2. Characteristics of Plan Construction A plan is a series of steps corresponding to possible actions in a domain that transform an initial state into a desired state. The need for good planning arises in many situations, from determining the path for an autonomous vehicle, to deciding what sequence of tests to run to diagnose a patient’s illness. In each case, the planning technique must determine the sequence of intermediate steps required to move from the initial state to the goal state. Thus, planning can be characterized as a search of all possible combinations of steps, with the resulting plan being one that provides a way of getting from the initial state to the goal state. In most 42 From: AAAI Technical Report WS-93-04. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved.