ance capabilities to reach
a given goal position.
• When the rover
reaches a new area, the
final pose is subject to
dead reckoning errors.
This requires a data
uplink to Earth to
analyze the state and to
plan future operations.
Martian day (which is called “sol”), as
their power supply is the sun. However,
there are many contingent and economic issues which must be addressed:
• Activities must occur every sol in
order to maximize scientific return.
• The current technology in cognitive
systems does not allow complete
autonomy for rovers on remote planets.
• The distance between Earth and Mars
precludes direct tele-operation because
of limited bandwidth and temporal lag.
Therefore, plans must be submitted
from Earth and executed on Mars
without direct supervision from Earth.
These issues lead to a
well-defined sequence of
activities which must be coordinated
between Earth and Mars: Earth-side processing and filtering of data downlinked
from Mars; analyzing the current rover
state and its surrounding terrain for possible hazards; locating traversable areas
and features of interests; and finally,
uplinking new commands every sol.
From the rover’s side, this implies
the execution of the received plan
schedule (i.e., a sequence of navigation
and scientific activities), the recording
of the relevant information, and then
the communication with Earth. Those
activities are outlined next:
• Rovers are semi-autonomous between
communication windows, thus requiring
at least self-localization and hazard avoid-
• Data analysis (Earth). Images and
rover data are received from Mars. This
step involves browsing images taken
from the various cameras on board the
rover (i.e., to examine the surrounding
terrain and to detect hazards and areas
of interest) and analyzing numerical
values originating from sensors.
Images are first displayed and
merged together, thus producing environment panoramas (see Figure 1
again). These images are used to
integrate scientific activity plans, which
act as guidelines for planning
approaches to targets of interest.
In practice, scientists ask rover
operators to move the vehicle according to what they see in image panoramas. In order to execute a plan, rovers
must be fully operational and operators
need to visualize the rover within the
terrain and analyze interactions with it.
First, the rover status is checked:
telemetry provided by such data
channels as the suspension and
the steering angles and internally
computed quantities (i.e., the robot
pose) are visualized and possible
dangerous situations are inferred.
Second, a 3D model of the rover
environment is built from camera
images (see Figure 3). This is usually
achieved using stereo vision. Stereo
image pairs are processed separately,
and then the left and the right images
are correlated to find matching pixels.
Next, the disparity of each matched pair
is computed, thus retrieving an estimated distance of the feature pixel from the
stereo camera axis. As a result, a full 3D
model of the rover surroundings is built.
• Plan generation (Earth). Once a clear
model of the rover environment is
known, it is possible to specify
command sequences for future
activities. Between communication
windows, rovers are autonomous in
that they operate while out of contact
with ground controllers. Roughly
speaking, a mission is composed by a
collection of actions to be performed
in sequence. In general, there is an
interleaved sequence of navigation and
science activity actions.
• Driving the rover (Earth). Using an
interactive 3D visualization software, it
is possible to specify desired rover positions in various ways. In general, this
can be achieved using point-and-click
50 SERVO 01.2008