A Hybrid Architecture for Real-Time Mixed-Initiative Planning and Control

Steven W. Mitchell

In many mission critical applications current technology is inadequate for fully automatic planning and control. In these applications society insists that planning and control be exercised by human minds. However, many such applications lie at the brittle edge of human capabilities. This has lead to serious incidents such as those involving the USS Stark and the USS Vincennes in the Persian Gulf. In domains like these where full automation is unacceptable and purely human operation is inadequate, a promising approach is one which combines the strengths of humans and computers.

This paper describes one architecture for addressing this challenge. Interacting with human domain experts in a mixed initiative mode it combines elements of case-based and model-based reasoning in a hierarchical task network decomposition planner to generate plans, and uses multivariate utility theory to evaluate the plans. The architecture includes real-time monitoring of plan execution, and automatic replanning for plan failure or significant changes in the environment. The planner has been implemented in C and C++, and used as the Tactical Response Planner for the DARPA Ship Systems Automation (SSA) program.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.