Information Technology Reference
In-Depth Information
the physical domain of the robotic system [2]. An analogy emerges here with
Toll-Like Receptors (TLRs) as sensors of potentially problematic signals within
the body. Such signals are known as Pathogen Associated Molecular Patterns
(PAMPs) [9]. In robotic systems simple sensors capable of detecting problematic
circumstances (eg. “motor3 overheating”) can often be used locally to help reme-
diate the problem without recourse to high-level software and control systems.
This is directly analogous to the types of action taken by innate immune sys-
tem components (such as macrophages) endowed with TLRs. Diculties arise
in engineering complex robotic systems (or other electro-mechanical systems)
which attempt to integrate the input from large numbers of such local sensing
and remediation devices into high-level control systems. It rapidly becomes im-
possible to predict all possible combinations of problem and remediation action,
and computationally expensive to process all this information in the high-level
controller. A number of approaches to robot control have addressed this problem
with varying degrees of success, the best known being [1]. The notion of artificial
inflammation allows the integration of information about low-level response pat-
terns into a small number of global signals which represent the “state of health”
of the system throughout time. These simple inflammatory signals can then be
used via schemes such as neuro-endocrine control [7,8] to modulate high-level
control systems appropriately.
The representation of the states of the robotic system using Kohonen's Self-
Organizing Maps (SOM) [6] allows the sources of the inflammation to be localized
within individual nodes in order to both diagnose problems at intermediate levels
(eg. “motor compartment 2 overheating”) and to allow higher-level remediation
to be appropriately targeted on the components that directly affect the inflamed
parts of the robot.
A description of the physiology of a robot follows, including the analogy drawn
from the innate immune system. Next, a step-by-step description of the model is
used to show exactly how it works both in this specific case and how the scheme
works in general. A proof of concept experiment is described, supported by the
results obtained and a commentary on what the results show. This is followed
by some conclusions, including advantages and disadvantages of the proposed
model.
2
Robot Physiology
In general a robot is a complex system made up of numerous interacting com-
ponents that can fail or malfunction alone as well as in combination. Typical
components also include automatic damage protection functions and circuits
such as locally switched cooling fans and automatic overheat cut-outs. Analo-
gies between such components and the innate immune system are presented here.
Firstly the function of TLRs in the innate immune system is the detection of
PAMPs. In a robot the proprioceptive sensors which monitor the state of the
robot can be considered to be analogous to TLRs. For example a temperature
sensor, measuring the temperature of a motor within a robot might have a TLR
