INTELLIGENT NAVIGATION FOR AUTONOMOUS ROBOTS USING
DYNAMIC VISION
C.Hock; E.D.Dickmanns
Universität der Bundeswehr München
W.Heisenberg Weg 39, D-8014 Neubiberg, FR Germany
Abstract
With the autonomous road vehicle VaMoRs behavioral
competences have been developed over the last decade for
visually guided longitudinal and lateral road following
including obstacle avoidance; these methods are numeri-
cally very efficient and locally adequate. They do not
allow global navigation. With the autonomously guided
vehicle ATHENE for transportation tasks on the factory
floor, indoor landmark navigation has been demonstrated
exploiting the same 4D-approach to dynamic machine
vision.
Combining the results of both application areas, a very
flexible and powerful intelligent navigation scheme is
achieved. The background and the basic features of this
new method are discussed.
Key words:
Navigation, Landmarks, Autonomous Robots, Dy-
namic Vision, Data Fusion.
1. Introduction
Route planning and visual guidance of vehicles has been
a subject of research in artificial intelligence for a long
time. The remote sensing capability of vision allows an
agent to orient itself relative to the environment and to
other objects up to relatively large distances.
In well developed road networks the capability to per-
form complex missions, clearly has three essential com-
ponents: 1. safe movement along the road disregarding
navigational aspects in the large (so-called cruise phases),
2. orientation on the mission scale and taking proper
navigational decisions when required, and 3. the capability
of implementing navigational maneuvers from the pre-
vious to the following cruise section.
The first task has been solved and demonstrated with
VaMoRs, a 5-ton van with proper sensing and actuation
capabilities, extending recursive estimation techniques to
image sequence processing with the 4D-approach [Dick-
manns, Zapp 86, 87; Dickmanns, Christians 89, 91; Dick-
manns, Graefe 88; Dickmanns, Mysliwetz 92]. Lane
following, convoy driving , stop-&-go in a traffic jam, and
lane changing, all have been demonstrated in the frame-
work of the EUREKA-project PROMETHEUS with the
"Common European Demonstrator-3' VITA of our in-
dustrial partner Daimler-Benz. Obstacles may be detected
at ranges up to 100 m and proper reactions are triggered
through situation assessment and feed-forward or feed-
back control actuations.
Task two has been tackled in our group first for guiding
vehicles on the factory floor [Hock 91]. If a flexible
scheme requiring little hardware installations is being
looked for, visual landmark navigation is the way to go;
the least expensive approach would be that well discern-
ible feature groupings already present in the environment
may serve as landmarks and are sufficient for reliable
recognition of the actual vehicle position. This is exactly
how humans and animals tend to find their way around,
even when due to changing lighting conditions and annual
seasons the appearance of landmarks changes systemati-
cally. The 4D-approach integrating temporal aspects right
from the beginning is well suited for realising this scheme
efficiently. It requires memory and knowledge processing
onboard the system. ;
The third task mentioned above also has been tackled
successfully: On the Autobahn, navigation is simply done
by proper lane changing and lane following; up to now,
the trigger impulse had to come from the human operator.
However, it is relatively simple to achieve full autonomy
once the capability of traffic sign recognition can be
incorporated. On normal roads, the capability of recogniz-
ing crossroads as landmarks and of turning off is the
behavioral competence required; this is being worked at
[Müller 92].
The term ’autonomous robot’ seems to be surprising at
first sight, since 'robot" per se is an autonomous device by
definition. Most of the industrial robots, however, still
have a link between a human operator and the machine.
The robot follows a predefined program with no choice of
making own decisions. A large amount of research work
in the field of robotics is devoted to reducing the need for
information exchange between man and machine. A nec-
essary step towards autonomy is to provide intelligence
within the onboard devices. Autonomous operation is then
determined by the intelligence of the machine.
The dictionary [Hornby 78] explains the word ’intel-
ligence' as the "power of perceiving, learning, under-
standing, knowing, mental ability'. Perception and
understanding of the operational environment for mobile
robots are the main aspects of research work performed at
UniBwM over the last decade.