Doman’s Inclined Floor Method for Early Motor Organization Simulated with a Four Neurons Robot - Foundations on Natural and Artificial Computation - page 116

Glenn Doman suggested placing the children on a ramp in a belly-down posi-
tion. The inclination of the ramp should be increased up to the point where the
movements of the child make his/her body slip on the ramp. For better results,
parents are suggested to shout or clap when the child successfully moves down
the ramp. Black and white stripes painted on the ramp contributes to improve
even more the performance of the method, also helping to stimulate children
sight.
Before placing the robot on the inclined floor its four artificial neurons fire
in a random way, and the wheels (controlled by these neurons) move also ran-
domly. Once the robot is placed on the ramp, it slides a little when it moves
downward (sliding does not take place when the robot moves upward). Sliding
makes the sensory input of the robot being highly activated and, since sliding
happens during downward movements, downward movements produce a higher
sensory input. In the case of the light sensor, the higher activation takes place
because the sensor identifies more transitions between white and black stripes
when sliding downward than when moving upwards. In the case of the ultra-
sonic detector it was modified to calculate accelerations, and accelerations are
also more intense when the movement of the robot is helped by gravity, i.e when
it slides downwards. The only stimulus in which gravity is not relevant is sound,
which in our case is produced by a maraca. The maraca is played when the robot
performs a correct movement.
Our working hypothesis is that, when a random sequence of neurons is such
that makes the robot goes downwards, the subsequent sensory stimuli contribute
to reinforce the weights between these neurons. In this way, different chains of
neurons linked by strengthened weights are formed, each chain representing a
specific way the robot goes downwards. The more intense the sensory stimuli,
the stronger are the connections in a chain of neurons.
To evaluate these hypotheses we counted the number of iterations (each iter-
ation corresponding to each movement of the robot) until the robot learns to go
downward under several arrangements of its environment. For example, a spe-
cific arrangement of the environment is the robot placed on the ramp (so that
it experiments body acceleration) without sound and without stripes painted on
the ramp (the behavior of the robot with some of these arrangements can be seen
on youtube by placing the authors' family names in the youtube searching tool).
We consider that the robot has learned when it moves in the same direction
during five consecutive iterations.
In Table 1 we show different arrangements of the robot environment in which
different types of sensory stimuli were used. We performed 15 tests for each
configuration. In each test we counted the number of iterations that it takes for
the robot to learn. The mean and standard deviation of the 15 tests of each
configuration are also shown in Table 1.

4 Practical Implications

According to Table 1, when two stimuli are combined, the number of iterations
it takes for the robot to learn is less than when a single stimulus appears alone.