Page 106 - Dynamic Vision for Perception and Control of Motion

3 Subjects and Subject Classes
90

tuators for processors on higher system levels. On the contrary, it is more likely
that after abstract decision-making, there will be several processors in the down-
link chain to the actuators. To achieve efficient system architectures, the question
then is which level should be assigned which task. Here, it is assumed that (as in
the EMS–implementation for VaMoRs and VaMP, see Figure 14.7), a PC-type
processor forms the interface between the perception- and evaluation level (PEL),
on one hand, and specific microprocessors for actuator control, on the other hand.
This processor has direct access to conventional measurement data and can close
loops from measurements to actuator output with minimal time delay.
The control process has to know what to do with the symbolic commands com-
ing from the PEL for implementing basic strategic decisions, taking the actual state
of the vehicle into account. It has more up-to-date information available on local
aspects and should, therefore, not be forced to work as a slave, but should have the
freedom to choose how to optimally achieve the goals set by the strategic decision
received from the PEL. For example, quick reactions to unforeseen perturbations
should be performed under the subject’s responsibility. Of course, these cases have
to be communicated back to the higher levels for more thorough and in-depth
evaluation.
It is on this level that all control time histories for standard maneuvers and all
feedback laws for regulation of desired states have to be decided in detail. This is
the usual task of controller design and of proper triggering in systems dynamics. In
Figure 3.17, this is represented by the lower level shown for longitudinal control.

3.4.5 Dynamic Effects in Road Vehicle Guidance

Due to the relatively long delay times associated with visual scene interpretation it
is important for instant correct appreciation of newly developing situations that two
facts mentioned above already are taken into account: First, inertial sensing allows
immediate perception of effects of perturbations onto the own body. It also imme-
diately reflects actual control implementation in most degrees of freedom. Second,
exploiting the dynamical models in connection with measured control outputs, ex-
pectations for state variable time histories can be computed. Comparing these to
actually measured or observed ones allows checking the correctness of conditions
for which the behavioral decisions have been made. If discrepancies exceed thresh-
old values, careful and attentive checking of the developing states may help avoid-
ing dangerous situations.
A typical example is a braking action on a winter road. In response to a com-
manded brake pressure with steering angle zero, a certain deceleration level with
no rotations around the longitudinal and the vertical axes are expected. There will
be a small pitching motion due to the distance between the points where forces act
(see Figure 3.9 above). With body suspension by springs and dampers, a second-
order (oscillatory or critically damped) rotational motion can be expected. Very of-
ten in winter, road conditions are not homogeneous for all wheels. Assume that the
wheels on one side move on snow or ice while on the other side the wheels run on
asphalt (MacAdam, concrete). This yields different friction coefficients and thus
different braking forces on both sides of the vehicle. Since total friction has de-