Information Technology Reference
In-Depth Information
examples can be found in the still timid robotic presence within the shipbuilding
sector in tasks like hull water-blasting [1] or welding [2]. However, automation of
these types of dynamic and unstructured industrial environments is in an early
stage, and, although robots are becoming more autonomous and flexible, in prac-
tice they are still designed for single tasks, with improved adaptive capabilities,
but with no capacity at all in terms of reconfiguration.
It is with the objective of adding more flexibility to the systems and allowing
them to be able to adapt to different tasks in different ways that multicomponent
robotic systems (MCR) come into play. Multicomponent robotic systems present
several features like scalability, fault tolerance, low cost of manufacturing and
maintenance or reconfiguration simplicity that make them highly suitable for
the kind of environments considered here. According to [3] MCR systems can
be classified into three main types depending on the degree of coupling among
robots: Uncoupled Distributed Systems, Modular Systems and Linked System.
Here we are interested the second type, modular systems. These systems are
based on the aggregation of modules with limited individual capabilities that,
through rigid physical couplings between components, can produce robotic units
with different physical and functional properties.
Developing a modular MCR system implies solving two problems: the me-
chanical and electrical design of the individual modules and the development of
a procedure for their aggregation in order to obtain the final morphology and
control structure for a particular task or set of tasks. There are several examples
in the literature of modular robot architectures such as polybot [4], M-TRAN [5]
or CONRO [6]. Most of these architectures have been designed as homogeneous
sets of modules [12][5][8] and with a scientific purpose of exploring the domain
and obtaining some basic principles, seldom taking into account the real con-
straints of industrial operation. An exception in this line is the work of Shen et
al. [8] [7] with the Superbot system. Here, the modules that make up the robotic
unit are an improved version of those in the M-TRAN and CONRO systems.
They have been designed for unsupervised real environment operation, resisting
abrasion and physical impacts, and including enhanced sensing and communi-
cations capabilities. This modular system is capable of climbing a rope or going
up stairs.
In terms of obtaining the morphology, deciding on what modules and how
they are connected, and the control system for the modular robot to perform a
given task or set of tasks, this has usually been done by hand. Obviously this
limits the applicability of the robots in reality as all their configurations would
have to be pre-designed and, in most cases, implemented by a human operator.
The reason for is this the surprisingly high dimensionality and complexity of the
search spaces that are induced by the possible robot morphologies as well as how
deceptive these spaces are, making most automatic approaches basically perform
as a random search. Several authors have applied evolutionary techniques to
solve in very simple instances this automatic design problem with successful
results [9] [10]. But the main problem we have found in these approaches is that
the evolutionary algorithm that is typically applied has not been adapted for
Search WWH ::
Custom Search