Information Technology Reference
In-Depth Information
3.5
Spreading Inflammation
The clustering effect of the SOM offers a way of dealing with local as well as more
widespread problems in a way which is analogous to inflammation. The robot
is in a stable/homeostatic state if all its actuators are working correctly. Once
problems start to occur, the nodes whichbecomeactivatedwithintheSOMfall
outside the cluster of the stable behaviour. Once in such an unstable state the
artificial innate immune system first deals with the problem locally at the level
of TLRs. In case this local prevention does not return the robot to a stable state
within a short period of time, inflammation starts to spread to neighbouring
nodes of the current state node. This way the system deals with the problem by
performing similar, yet slightly different responses, until the problem is rectified
and the robot is returned to a stable state (a node within the SOM is activated
which belongs to the cluster of stable/homeostatic behaviour).
4
Proof of Principle
A proof of principle implementation has been developed to demonstrate the key
features of the operation of the model as described above. The model contains
a small number of TLRs and uses inflammation responses generated by them to
modify behaviour of a very simple high level control system. The inflammation
response is integrated across the system and is decayed in the manner indicated
above. Simple physical models of heating and cooling of motors are included in
the simulation. The SOM component is not currently integrated into the system,
but the vectors representing the system state were collected during the execution
of the model and were used to train an SOM to prove the principle. This imple-
mentation has been performed as a simulation containing the important parts
of the robot's functionality. The following results were obtained, supporting the
proposed principle and its viability in a future physical system implementation.
4.1
Description of the Model
The simulated robot has two motor compartments: one for the front two wheels
and one for the rear two wheels. Each wheel has a separate motor as is common
in all-terrain robots. Each compartment also has a single cooling fan which is
responsible for cooling the pair of motors in that compartment. Each motor has a
TLR associated with it which monitors the motor's temperature. Each TLR has
three possible states. The “normal” state is that the motor is enabled and the
fan is switched off. When the motor reaches a predefined threshold temperature
the TLR will activate and switch on the fan in that compartment. If the motor
reaches a second, higher threshold temperature which endangers the motor then
the TLR will activate a thermal cut-out which cuts all current to the motor in
question in order to allow it to cool. This disables the motor and thus deprives the
high-level control system and the robot as a whole of the use of that motor. The
simulation ensures that the temperature of the motors increases proportionally
to the current passing through it. The motor model also includes a simplistic but
