Abstract
Over the last years, the Artificial Intelligence area is undergoing a great
advance. It is widely applied in many areas, one of which is the area of
real-time problems. In these problems, the logic of the solution calculus is
important, but also the time when the solutions are obtained.
The fusion of Artificial Intelligence and Real-Time areas is useful, because
the Artificial Intelligence provides new possibilities to the Real-Time
systems. New possibilities such as the flexibility for solving problems in
complex and dynamic environments. However, this approach has shown
important difficulties.
Mainly, the Real-Time systems have temporal requirements (they usually
require predictable response times) that are not usual in Artificial
Intelligence techniques.
One of the ways to solve this problem is the development of software
architectures. These software architectures are used to design intelligent
agents that work in real-time environments. These architectures have
several mechanisms to allow to the build agents to work in real-time
environments offering reactive behavior (to fulfill the temporal
requirements) and deliberative behavior (that make use of Artificial
Intelligence techniques to obtain better results).
One of these architectures is ARTIS. This architecture uses a two-level
scheduling of its tasks to fulfill this objectives. A first-level scheduler
guarantees to obtain the solutions within the hard-time limits. A secondlevel
scheduler controls the components used to improve the result quality.
The methods shown in this work are centered in the second-level scheduler.
Two new heuristics, SSS (Slack-slide scheduling) and SSSM (SSS with
Memory) has been developed. These heuristics make a more precise use of
the architecture characteristics and the components to be scheduled. This
way, they can obtain better results qualities than other applicable
heuristics.
These heuristics are able to handle several types of algorithms that the
architecture ARTIS uses (progressive refinement and multiple methods).
One of them, the SSSM heuristic , uses the resources of the architecture to
take advantage of previous results to improve the system efficiency.
Finally, the memory shows statistical tests to verify the heuristic
characteristics and two real application to show its viability.