Dependability of on-line optimization techniques in real-time applications

Abstract

Real-time problem solvers require dependable, real-time search algorithms to meet task deadlines and to predict deadline violations. Presently it is difficult for existing real-time search algorithms to search and execute the solution by the deadline and to make deadline violation prediction. We introduce a real-time search algorithm called Self-Adjusting Real-Time Search (SARTS). Given a timing constraint, SARTS adjusts itself based on the remaining time to deadline and allocates the planning time. As the timing constraints are relaxed, it will continue to improve its solutions progressively. The algorithm is able to predict deadline violations. Theoretical analyses and experimental results reveal that, compared to the existing techniques, SARTS demonstrates a higher degree of predictability and a higher deadline compliance ability.