Emily Lakdawalla
Page 29
230 The Mast, Engineering Cameras, Navigation, and Hazard Avoidance
6.5 ROVER DRIVING
The rover drivers plan rover motion using a variety of local coordinate systems. They can
instruct the rover to use various amounts of artificial intelligence to complete a drive. From less to more autonomous, the rover driving modes include blind driving, visual odometry
(“visodom”), and autonomous navigation (“autonav”). Another mode, “guarded motion,”
is a hybrid of visodom and autonav. Rover autonomy has a trade-off, because the greater
the rover computing power required to drive safely, the slower the rover moves. To drive
for distance, a drive may include segments of blind driving, then visodom, then autonav
until reaching a time limit.
6.5.1 Coordinate systems
Placing the rover’s scientific observations in geographic context is crucial to interpreting them. The rover has inertial measurement units to dead-reckon its position and orientation.
Ideally, all rover measurements would be tied precisely to a latitude/longitude/elevation
spatial frame, but this can’t happen automatically because of imprecise instantaneous
knowledge of the rover’s location.
The quality of the rover’s position information degrades with time, for two reasons.
First, the wheels slip. This means that the amount of distance the rover has traveled is
never quite the same as the distance commanded. If wheels on one side slip more than
those on the other side, slip results in unexpected rotation as well as distance. And second, the bumping and jostling of the rover as it travels over rough terrain accelerates the inertial measurement units in ways that can be incorrectly interpreted as distance traveled.
To help manage the uncertainty in rover position and to compartmentalize the errors,
the mission keeps track of several different spatial reference frames.6 The two most commonly used ones are the rover frame and the site frame. The rover frame is fixed relative
to the rover. The rover frame origin is at a spot on the ground between the middle wheels
(assuming the rover is perfectly level). In the rover frame, +X is forward, +Y is to the right, and +Z is down. A site frame has its origin at a fixed point on the surface of Mars. The
rover performs operations like camera pointing, arm activities, and drives relative to the
site frame. The site frame has +X pointing north, +Y pointing east, and +Z pointing down-
ward in a direction perpendicular to the map. Over time, error accumulates in the rover’s
reckoning of its motion relative to the site origin. Periodically, the team declares a new site origin and increments the site number. By keeping careful track of where measurements
were made in the rover frame, and precisely determining the geographic location of each
site frame, science measurements can be precisely geolocated.
When the mission declares a new site origin, the spatial position is determined by com-
paring Navcam photos to orbital image data, but it’s harder to precisely identify the rover’s orientation in space. Curiosity’s inertial measurement units provide continuously up-to-date pitch (front-to-back tilt) and roll (side-to-side tilt) information, but the rover’s
6 The various reference frames are described in detail in Alexander and Deen (2015).
6.5 Rover Driving 231
Figure 6.6. A typical right Navcam image of the Sun, taken to support a new site frame declared after a drive on sol 324. The horizontal line is pixel bleeding caused by overexpo-sure. Image NRB_426264304EDR_F0060864SAPP07612M. NASA/JPL-Caltech.
knowledge of its yaw (compass orientation) degrades over time. Curiosity periodically
updates its yaw knowledge by shooting a mid- to late-afternoon photo of the Sun with the
right Navcam. Even with pixel bleeding, the rover can identify the location of the Sun
precisely enough to identify its yaw relative to the local coordinate system (Figure 6.6).
6.5.2 Driving modes
6.5.2.1 Blind driving
In a blind drive, the rover doesn’t employ any onboard intelligence to look at the landscape during the drive. Instead, the rover planners examine a 3D model of the landscape or “terrain mesh” calculated from Navcam and Hazcam images, and command the rover to roll
its wheels a certain distance, turn through a specific number of degrees, and so on. The
lengths of blind drives are limited to the distance that the rover can see well enough with the Navcams to develop a terrain mesh, usually no more than 50 meters. Blind drives can
be longer than 50 meters if the terrain slopes upward and is benign. If the terrain is slippery (as it may be if it’s sandy or sloping), blind driving can be inaccurate. Blind driving is the fastest mode, achieving speeds of roughly 100 meters per hour.
232 The Mast, Engineering Cameras, Navigation, and Hazard Avoidance
When executing a blind drive, the rover doesn’t perform any checks to make sure it is
on course. It does always perform checks to make sure that the mobility system is operat-
ing within safety limits, and will stop the drive short if (for example) there is too much tilt or too much resistance to the motion of a wheel. The rover planners may set these limits
differently for each and every drive: a drive over smooth terrain should result in little rover tilt, so they’ll set tilt limits lower than they would for a drive over rockier terrain.
6.5.2.2 Visual odometry
Visual odometry, or “visodom”, helps the rover maintain the course that the rover driv-
ers set. During a drive, the rover looks to the side with its Navcams, taking stereo images at specified intervals (ranging from 50 to 150 centimeters). The rover computer compares pairs of images, matching features between image pairs, to determine how far the
rover actually moved. The rover can then re-plan its path based upon its determination
of how far it judges it has actually traveled, or can stop its travel if it is not making sufficient progress due to wheel slippage. Visual odometry slows the rover to roughly 50
meters per hour.
6.5.2.3 Autonomous navigation and guarded motion
Autonomous navigation, or “autonav”, is an even more sophisticated autonomous driving
capability that allows the rover to drive beyond its terrain mesh. The rover drivers identify a goal, specified as a position in the local site frame coordinate system. The rover moves a short distance of 50 to 150 centimeters. It snaps Hazcam images and processes them into
3D information to update the terrain mesh. It identifies obstacles exceeding 50 centimeters in height and slopes steeper than 20°. The rover charts the “traversability” of a square of nearby terrain extending 5 meters around the rover, divided into a 20-centimeter grid. Each grid cell is assigned a “goodness” and “certainty” estimate that rolls together the rover’s determination of the safety of that patch of terrain. The rover fits models of itself into this map to find the safest path. It rolls forward by another increment of 50 to 150 centimeters depending on how safe it perceives the terrain to be, then repeats the Hazcam imaging and
evaluation process. Because of all the calculation, autonav is slow: a top speed of about 50
centimeters per minute, or about 30 meters per hour.
A related form of driving is “guarded motion,” where the rover planners give the rover
a specific path to follow using visual odometry, but then instruct the rover to use autonav to verify that the path is indeed safe as it moves forward.
The use of autonav was ended following discovery of the wheel degradation problem
(see section 4.6.4); mitigating wheel damage required rover planners to avoid hazardous terrain on a scale finer than the 20-centimeter grid used by autonav. It was re-enabled as of sol 1780,
and planners have discretion to choose whether the local terrain is benign enough to enable autonav.
6.6 References 233
6.5.2.4 Multi-sol driving
When Curiosity landed, it could not save the terrain maps generated one sol and use them
on the next sol. As part of a set of improvements included in flight software version R.11, implemented on sol 484, engineers added the ability to save on-board terrain maps during
sleep to enable the rover to use the same one to continue a drive the next day, increasing
the drive distances achieved during traverse periods.
6.6 REFERENCES
Alexander D and Deen R (2015) Mars Science Laboratory Project Software Interface
Specification: Camera & LIBS Experiment Data Record (EDR) and Reduced Data
Record (RDR) Data Products, version 3.5.
Kloos J L et al (2016) The first Martian year of cloud activity from Mars Science Laboratory (sol 0–800). Adv Space Res 57:1223–1240, DOI: 10.1016/j.asr.2015.12.040
Lemmon M T et al (2017) Dust devil activity at the Curiosity Mars rover field site. Paper
presented at the 48th Lunar and Planetary Science Conference, The Woodlands, Texas,
20–24 Mar 2017
Maki J et al (2012) The Mars Science Laboratory engineering cameras. Space Sci Rev
170:77–93, DOI: 10.1007/s11214-012-9882-4
Moores J E et al (2014) Update on MSL atmospheric monitoring movies sol 100–360.
Paper presented at the 45th Lunar and Planetary Science Conference, The Woodlands,
Texas, 17–21 Mar 2014
7
Curiosity’s Science Cameras
7.1 INTRODUCTION
Curiosity has five science cameras. The color Mastcams view the rover’s world in color at
two different resolutions. The Mars Hand Lens Imager (MAHLI, pronounced “Molly”) on
the turret at the end of the arm, is a wide-angle color camera that can be held close to a
target or perform distance imaging. The Mars Descent Imager (MARDI) is fixed to the
rover body, pointing down, with a view of the surface as it passes under the rover. Together, these three instruments are often referred to as the “MMM” cameras. They have common
detector and electronics and software design and differ only in their optics. Finally, there is the laser-equipped ChemCam, which measures elemental compositions of nearby rocks
and also possesses the camera with the highest angular resolution on the rover, the Remote
Micro-Imager (RMI). It will be described in Chapter 9 with the other composition analysis
instruments.
Figure 7.1 shows the locations of camera instruments and related hardware on the rover.
The engineering cameras (Navcams and Hazcams, section 6.3) serve science functions as well. They provide context for science observations and perform remote sensing science
observations, particularly atmospheric science. Table 7.1 compares all of Curiosity’s imaging capabilities.
7.2 MASTCAM
The Mastcam instrument consists of two camera heads located on the mast, an electronics
assembly located in the belly of the rover, and a calibration target on the rover deck. With the Mastcams, the science team investigates geomorphology, stratigraphy, and texture of
the landscape, rocks, and sediments around the rover. They also monitor atmospheric and
even astronomical phenomena. They support the rover’s engineering activities and provide
© Springer International Publishing AG, part of Springer Nature 2018
234
E. Lakdawalla, The Design and Engineering of Curiosity, Springer Praxis Books,
https://doi.org/10.1007/978-3-319-68146-7_7
7.2 Mastcam 235
Figure 7.1. Locations of camera instrument components on the rover, as well as some devices often imaged with Mastcams. Mastcam, Navcam, and ChemCam covers in top image were
used only during cruise and landing. Top image is cropped from the Gobabeb MAHLI self-
portrait mosaic, sol 1228. Bottom image taken at JPL during assembly. NASA/JPL- Caltech/
MSSS/Emily Lakdawalla.
236 Curiosity’s Science Cameras
ver
vement
MARDI
1600 × 1200
70–52
0.76
yes, with
ro
mo
arbitrary
no
0.66
420–690
Bayer color
vement
MAHLI
1600 × 1200
34.0–38.5
0.402–0.346
yes, with arm
mo
arbitrary
yes
arbitrary
420–690
Bayer color
RMI
1024 × 1024
1.3
0.022
no
–
yes
2.1
450–950
monochrome
ferent
MR
6.8 × 5.1
0.074
V
8 plus
Bayer
ye
ut with dif
ML
1600 × 1200
20 × 15
0.22
yes, b
resolution/FO
in each e
24.5
yes
1.9
395–1100
8 plus
Bayer
. as
av
s camer
N
1024 × 1024
45 × 45
0.82
yes
42.4
no
1.9
600–800
monochrome
RHaz
10
0.78
FHaz
1024 × 1024
124 × 124
2.1
yes
16.7
no
0.68
600–800
monochrome
el)
els)
ace (m)
Comparison of the capabilities of Curiosity’
ve surf
V (°)
V at center (mrad/pix
Table 7.1.
CCD Detector (pix
FO
IFO
Stereo?
Stereo
separation (cm)
Depth information from focal
depth?
Height abo
Spectral bandpass (nm)
Filters
7.2 Mastcam 237
Figure 7.2. Parts of the Mastcam instrument. Photos of the Mastcam-100 camera head and digital electronics assembly were taken at Malin Space Science Systems before their delivery to JPL for assembly. Bottom self-portrait taken at the John Klein drill site on sol 177 by MAHLI. Inset self-portrait showing the back of the camera heads and their wire harnesses taken at Okoruso drill site, sol 1338. NASA/JPL-Caltech/MSSS/Emily Lakdawalla.
context images for data from other science instruments. The Mastcams were built by
Malin Space Science Systems, San Diego, California. The principal investigator for the
Mastcam experiment is Michael Malin of Malin Space Science Systems.
The Mastcams differ from previous lander cameras in two significant ways. First,
nearly all Mastcam views are in full, human-vision-like color. Second, the two camera
“eyes” have different focal lengths, which makes stereo imaging more complex than for
previous missions. (Read section 1.5.8 for the history of the development of Mastcam that
238 Curiosity’s Science Cameras
Table 7.2. Mastcam facts.
Mastcam-34
Mastcam- 100
(Mastcam-L)
<
br /> (Mastcam-R)
Boresight height above bottom of wheels
1.97 m
Elevator actuator axis height above
1.91 m
bottom of wheels
Stereo separation
24.64 cm
FOV (horizontal 1600 pixels)
20.6°
6.8°
FOV (vertical 1200 pixels)
15°
5.1°
instantaneous field of view (IFOV)
218 μrad
74 μrad
Pixel scale at a distance of 2 meters
450 μm
150 μm
Pixel scale at a distance of 1 kilometer
22 cm
7.4 cm
focal ratio
f/8
f/10
effective focal length
34 mm
100 mm
in-focus range
0.4 m to infinity
1.6 m to
infinity
exposure range
0 to 838.8 s in 0.1 ms increments
video frame rate
5.9 to 7.7 fps at 720p (1280-by-720)
3.9 to 4.7 fps for full frame
led to the flight of a pair of Mastcams with different focal lengths.) The left Mastcam or
Mastcam-34 has shorter focal length, lower angular resolution, and wider field of view.
The right Mastcam or Mastcam-100 has longer focal length, higher angular resolution,
and narrower field of view.
7.2.1 How Mastcam works
7.2.1.1 Camera heads
The Mastcams are 2-megapixel color cameras with focusable lenses and filter wheels. 1
The heads contain electronics, a detector, a filter wheel assembly, a focus mechanism, and
a sunshade/baffle that also serves as a mount (Figure 7.2). Each head contains two stepper motors, one to drive the filter wheel and one to drive the focus mechanism. The two
Mastcams have boresights separated by 24.64 centimeters, and they are angled inward by
2.5° (1.25° each) in order to ensure that the smaller field of view of the Mastcam-100 is
entirely contained within the wider field of view of the Mastcam-34 for any target located
farther than 1.4 meters away from the rover. The boresights cross at a distance 2.8 meters
1 Prior to landing, there was no peer-reviewed paper describing Mastcam or MARDI. Mastcam was described in two Lunar and Planetary Science Conference abstracts: Malin et al. (2010) and Bell