Abstract


From the last century the human being has persecuted the dream to construct an independent robot and with a great availability in his operation able to make intelligent tasks similar to those of the humans to make simpler and pleasant life of these. Moved by the eagerness to construct an independent robot, the first movable projects of robots date from 70 years' s although the techniques available at those moments did not allow to reach the perspective created initially. S was at the beginning of the 80' when the investigation in this field returned to take special height. At the present time, without to have reached the objective to construct robots movable totally independent personal, great advances in this field have been obtained. In most one is prototypes elaborated by universities that happen to be part of the industry in robotics for their later commercialization. Some companies like Honda that has destined great sums to the investigation in this field at the moment already have developed robots very complex which they are being used in the automotive industry, in the area of the construction, the mining, aeronautics, in hides, in agriculture, the medical sector, exploration in the Mars planet, etc.
 

The developments in robotics indicate that in the next decade all a new industry referring to this new technology will exist, since robots will be articles of common use as it has happened at the moment to the computers. This situation entails to the necessity to construct robots available and highly reliable, reason why it is imperative to include in his design a fault tolerant system.


This work is centred in tolerating the failures at level hardware and software in the system of control of a movable robot, in such a way that the supervision, detection and recovery of failures independently implement to the system of control and the platform in which the robot is developed. This is obtained by means of the design of a fault tolerant architecture implemented with a System Multiagente (MAS). This system integrates a group of agents in charge of the detection and diagnosis of failures.

The fault tolerant architecture is integrated by two types of agents mainly those that are in charge to detect and to recover failures at software level (tasks) and the ones in charge to tolerate the failures at level hardware (sensorial, actuators, memories, controller of network, microcontroller, etc.). These fault tolerant agents execute the fault tolerant mechanisms of a very simple way making one storing of their characteristics that are the intercommunication and cooperation among them, thus being able: to detect, to isolate, to reshape and to try to recover to a component before failures (at level hardware and software) that appears during the operation of robot. 


In order to be able to develop efficiently propose the fault tolerant architecture it was necessary to modify the architecture of control at denominated level software integrated 3+ in the robot, as well as the physical architecture (distributed composed by nodes, where to each node it at the most connects 2 devices to him or of entrance and/or exit, each node counts on a microcontroller, and his tasks of control, navigation and planning).
 

The SMA that constitutes propose the tolerant architecture failures, was designed using the MaSE methodology (Multi-Agent Systems Software Engineering) [DeLoach 2001] since their contraction this made under defined mathematical models good. The use this methodology to model the SMA was very useful since it served to us as guide in the development of this system, in addition allowed us to easily make the transition of the phase of analysis to the phase of design, basing to us on the initial requirements of the system of a ascertainable comprehensible way, without ambiguities and contradictions.
 

It was required to make the analysis of used the software structures in robots movable in such a way that denominated 3+ were selected and I implement the system developed in this work Doing a modification in the third layer this software architecture, with the purpose of connecting our fault tolerant SMA, obtaining an efficient operation after making the pertinent tests in the simulator.


The theory of the diffuse sets like a tool for the diagnosis of the system by means of a tree of diffuse failures is investigated. The objective is to diagnose failures in components by means of the observation of diffuse symptoms being used the information contained in a tree of failures. For the solution of the problem a procedure in two stages is followed. In the first step, the causal reasoning is used to diagnose the failure ways that consist of making the minimum cut to the basic events of the tree of failures, the observation of the driven doors are treated like symptoms. In the second step, the particular components are identified that have failed based on the diagnosed ways of failure. In order to make this second step, the solution is by means of diffuse a relational equation to = V-product (alpha x of the t of S/sup) connecting the way of failure to the basic events derived from x. With this method, the equations of the diagnosis can be generated and be solved symmetrically in terms of basic events of the tree.

As last step design a simulator under the platform of JADE with the purpose of proving and of validating the here propose architecture.
 





1