Chapter 3 Self-Organizing Robots: An Outline
Abstract:
The past two decades have seen a drastic change of paradigms in the
control of autonomous robots. While in the strong AI paradigm the robot is
seen as a machine under total control, the new or embodied AI recognizes the
role of the body as an equal partner in the control process. The exploitation
of the specific properties of the body, the so called morphological
computation, not only reduces the computational load to the controller in
specific tasks like walking or swimming but also leads to smoother and more
natural motions of the robot. Although quite successful, the method seems
restricted to specific tasks and requires an inspired designer for the “morphological computer”, instead of an intelligent programmer.
Self-Organization (SO) may take us one step further on this way to embodied control.
Viewing a robot in its environment as a complex dynamical system, SO can help to let
highly coordinated and low dimensional modes emerge in the coupled system
of brain, body and environment.
In this way the robot may find out by itself
what its bodily affordances are and
focuses only in a second step on the exploitation of the emerging motion patterns.
The book describes two routes to the realization of this scenario,
one based on the dynamical system approach leading to the minimization
of the so-called time loop error,
which is the basis for the Homeokinesis,
and a second one based on information theory.
The present chapter will give a short outline of the method based on the dynamical systems approach.
3.1 Dominated by Embodiment: The Barrel
In the following we are going to discuss different control paradigms using the example of the Barrel.
This allows us on the one hand to demonstrate some features of the embodied AI (EAI) approach in a simple and transparent manner and on the other hand to outline the perspectives of SO for extending this approach to a wider field of applications.
3.1.1 Open Loop Control
| Video 3.1: The Barrel controlled by a periodic control signal with a phase shift of
π/2 between the internal axes. Starting with a very low frequency of the
controller signal, the frequency is doubled at time 2:15, 2:40, 3:05,
and 3:30. At 3:45 the Barrel is accelerated by a force (red dot)
but is seen to return rapidly to the original mode of behavior. The higher
frequencies very clearly reveal the difficulties of the motors to execute the
actions (periodic motions of the weights). |
3.3 Homeokinesis: Body Inspired Control
| Video 3.2: Behavior of the Barrel when starting in a situation where the center of
gravity of the Barrel is very low so that the situation is physically
stable. The homeokinetic learning is seen to destabilize the system quite rapidly (a few
seconds real time) and starts to move the Barrel. The dynamics of the
parameters of the model and the controller can be followed in the panels at
the left and right upper corners, respectively. The panels depict the course
of the parameters in a time window of 250 steps corresponding to about 10 sec.
After some time we obtain the stable rolling patterns.
Later on (at time 01:05) the Barrel was stopped by applying a physical force to
it (note the red dot which pulls the robot).
After releasing the force, the system recovers again in a very short time
and resumes the rolling patterns in a slightly modified form. |
| Video 3.3:
Creativity in unexpected situations: The Barrel was put into an upright
position by an external force about 10 seconds ago. Internal axes are
horizontal now so that the sensor values, the inclination of the internal axes, are zero apart from some small sensor noise.
The self-model does not get any reliable information in this situation so
that rapid forgetting sets in. This is counteracted by the controller
which increases weights so that any small perturbations are amplified. After
some time the motion of the internal weights become so strong that the
Barrel is tossed over.
The insets are the same as in Video 3.2.
Note that the scales of the panels change, in particular the model
parameters change by two orders of magnitude. The rapidly oscillating
parameters are the bias terms of both model and controller. |
| Video 3.4: Emergence of nontrivial modes. Out of the upright position there are different
modes that can emerge. In the video you see the emergence of a precession mode
which lasts for more than five minutes. Parameter and model dynamics is
depicted in the panels in the right and left upper corners (interchanged),
respectively. Note that the scales of the panels change, in particular the
model parameters change by two orders of magnitude. |
| Video 3.5: The emergence of nontrivial modes like in the barrel case are a common
phenomenon when using the general learning rules
for different agents. In the video you see a simulated humanoid robot. The joints are actuated by simulated servomotors.
Sensor values are the measured joint angles. There is no other information available to the robot. After the robot has fallen into a narrow pit, it develops different new motion patterns adapted to the new situation. After some time one often observes the emergence of climbing like behavior patterns.
The patterns, although they might help to get the robot out of this impasse,
emerge without any goals as a consequence of the sensitive but active interplay
between robot and the specific environment. |
| Video 3.6: Decay of nontrivial modes. One special feature of our approach is that modes
do not last forever, instead the concomitant learning of model and controller
most often leads to a slow change of the parameters so that the system leaves
the mode. In the present case, the precession mode, after lasting for about
five minutes decays spontaneously. The parameter change is most prominent in
the diagonal elements of both the model (left) and the controller (right
panel) depicted by the red and purple lines (which almost coincide). |
| Video 3.7: Emerging motion patterns are transient and depend decisively on the interaction with
the environment. Most complex patterns may emerge if the "environment"
is another robot of the same kind. The only sensors are the measured joint angles, like in Video 3.5
so that the two humanoids can only "feel" each other
by the mismatch between true (measured) and nominal joint angles resulting from the load on the joint.
Adherence is due to normal friction but essentially also from a failure of the ODE
physics engine, which after heavy collisions produces an unrealistic penetration effect which acts like a special gripping mechanism.
So, the "fighting" is a truly emerging phenomenon not expected before we saw it. |
Chapter 4 Guiding with Cross-Motor Couplings
Abstract:
Many robotic systems and their typical behaviors have a symmetric structure.
We will now propose a way to embed these symmetries as
constrains to the learning system, such that mainly desired behaviors emerge.
Starting from the guidance by teaching we introduced the concept of cross-motor coupling
that allow to specify abstract relations between motor channels.
First we studied simple pairwise symmetric relations and shaped the behavior
of the TwoWheeled robot to drive mostly straight through the coupling
between both motors.
Then we will consider a high-dimensional robot – the Armband
and demonstrate fast locomotion behaviors from scratch by guided self-organization.
| Video 4.1: Behavior of the Armband robot with cross-motor coupling and weak guidance (γS=0.001).
A slow locomotive behavior with different velocities is
exhibited. Explorative actions cause the
posture of the robot to vary in the course of time. |
| Video 4.2: Behavior of the Armband robot with cross-motor coupling and medium guidance (γS=0.003).
Comparable fast locomotive behavior emerges quickly and is persistent.
Nevertheless the velocity varies.
Only small exploratory actions are takes, such that the posture is mainly
constant. |
| Video 4.3: The behavior of the Armband robot with cross-motor coupling when the connections are changed.
The video start with a fast locomotive behavior to the left (k=1).
At time 5:00 the couplings are changed (k=0) and the robot slowly stops.
A period of probing actions follows until a reversed locomotion starts
to show up. |
This document was translated from LATEX by
HEVEA.