The goal of this project is to extend the multiagent system for robot navigation, developed in the project DPI2000-1352-C02-02, by means of a CBR (Case-Based Reasoning) Agent.
This agent will provide learning by experience capabilities to the navigation system. This new functionality will be based upon the detection (via vision and laser sensors) of environment situations that may affect the behaviour of the robot in such away that when confronted with a situation similar to previously experienced ones, the CBR Agent should be able to forsee the results of navigation actions based on what happened in the past similar situations.
We will focus our attention to those situations that lead to failures in reaching the navigation goal in order to avoid a new failure. However, we will also take into account those pairs of situation-action that lead to successfully reaching the goal.
To develop such CBR agent, we must deal with what can be called continuous CBR, that is to study how to represent in the case-base simbolic abstractions (high-level knowledge) of low-level continuous information coming from the sensor readings.
Another relevant aspect is that concerning the notion of similarity between situations and, in particular, how to measure these similarities. How to reuse a past successful solution is also a relevant aspect to be developed. It is worth noticing that there are extremely few existing efforts in CBR research dealing with continuous CBR. We, therefore, face an extremely important and interesting research problem.