Index Home About Blog
Date: 13 Sep 1993 23:31:00 GMT
From: Jordin Kare <>
Subject: Optical spacecraft navigation?

In article <> (Andre Spiegel) writes:
>  o  To what extent is optical navigation still being used? ..
>  o  How is it being done in detail? From what I read it seems there are
>     "coarse-tuning" and "fine-tuning" facilities.

Following up on several other replies...

JPL defines three classes of star attitude sensor:
	"Star Camera" detects images of stars
	"Star Tracker" includes a camera plus hardware/software that
		puts out some intermediate level of information, such
		as a list of star centroids in spacecraft coordinates
	"Stellar Compass" is a star tracker plus software that can
		identify stars and figure out the spacecraft attitude.

Most Earth-orbiting spacecraft use sun and Earth sensors only; Earth 
sensors are frequently far-IR sensors since they don't care how much of
the Earth is sunlit; they sense day and night limbs equally well.

Most older deep-space satellites and, e.g., astronomical observation
sats needing stellar references used various narrow-field-of-view
"Canopus trackers", scanned slits, etc. to update inertial (gyro)
references.  Some of these were extremely fine resolution (milli-arc-second)
using sophisticated optical systems.  Most required fairly accurate
initial positioning to be able to figure out where they were, and/or
extensive ground processing of data.

LLNL and a few other groups have developed wide-field star cameras and
hardware/software to make true stellar compasses.  LLNL's Star Tracker
(which is a stellar compass using JPL's terminology) is a 250-gram
(yes, 250 gram) CCD camera with a 28 x 40 degree field of view, 
plus a set of software that 1) identifies and centroids the brightest
starlike objects in the field, and 2) matches them to a list of stars
by using "similar triangles" of sets of 3 stars (which makes the tracker
immune to brightness errors or focal length changes).  The wide field
allows orientation anywhere in space (as long as the tracker can see stars)
using only about 600 bright stars.  The accuracy is about 100 microradians
on 2 axes and 300 microradians in roll.  

Some space probes have used optical navigation using science cameras to
determine position as well as orientation.  However, processing of images
and calculation of position was done on the ground.

There is currently much interest in _autonomous_ optical nav, which has 
never been done, in which the spacecraft itself figures out where it is, 
and therefore how to aim its cameras and/or change its trajectory.  
BMDO (ex SDIO) will be testing autonomous navigation algorithms on 
Clementine; the spacecraft will do all the processing needed to correct
its own trajectory to achieve an exact path past asteroid Geographos, but
will not have "permission" to fire its own thrusters; that will be by
ground command.  Future missions will, one hopes, be allowed more
capability.  Some future missions, e.g. Pluto Fast Flyby, pretty much
require autonomous optical nav to get the best possible trajectory.

	Jordin Kare

From: (Jordin Kare)
Subject: Re: Lunar Positioning System
Date: Sun, 16 Jul 2000 05:02:34 GMT

In article <>, (Henry Spencer)

> In article <8heo5m$t8o$>, gbaikie <>
> wrote:
> >>'d be better off with an
> >> automated star-tracker plus a tiltmeter to give a horizon reference.
> >> You wouldn't need any satellites for that.
> >
> >Wow, that's interesting. Day or nite and all you need is the know what
> >day and time it was.
> Yep, works fine.  And a Clementine star tracker is, if memory serves,
> about half a kilogram, although you need some computing power behind it.

300 grams.  And could be lighter today, especially if you provide a stable
base so you can integrate multiple frames, which would let you use an even
smaller lens.  The computing power required is actually quite modest; the
Clementine computer was around 20 MIPS IIRC, and could generate a new
attitude in 100 ms.

> >How accurate could this star-tracker be?
> The Clementine star trackers had a nominal accuracy of 150 microradians in
> pitch and yaw and 450 in roll (rotation around the tracker optical axis).
> Let's see, on the Moon a microradian is about 1.7m, so we're talking
> accuracy of about a kilometer.

Again, you have in some respects much less stringent requirements, and
technology has improved significantly (e.g., larger CCD's are readily
available).  Somewhat more massive star trackers routinely provide 10 urad
accuracy.  At a rough estimate, you could probably build a 300 gram, 1
watt sensor-plus-computer ("stellar compass") that would give you between
20 and 50 urad 1-sigma in pitch/yaw, one second after being set in a
stationary position. And assuming you can use a zenith view there should
be no strong reason to use the roll axis.  So 40-80 meter accuracy.

> Actually, the tiltmeter may well be the
> limiting factor.

Possibly, if you want a quick response time and a very compact sensor.
But given a few seconds and a few cm, I'd be surprised if you couldn't get
the same sort of accuracy.  After all, surveyors using not-very-complex
technology (bubble levels) have managed ~100 urad or better (a few cm per
km) on Earth.

Incidentally, while masscons might well affect the raw accuracy, they're a
static, fairly large-scale phenomenon and can therefore be compensated for
-- find an approximate position, look up the local masscon correction
factor, find exact position.  Not unlike correcting for magnetic deviation
on Earth.

Jordin Kare

From: (Jordin Kare)
Subject: Re: On-board instrument pointing systems?
Date: Sun, 16 Jul 2000 05:45:29 GMT

In article <8in8in$43a$>, Anttila Matti <> wrote:

> I am gathering information about on-board instrument pointing systems,
> especially star tracking systems and other optical systems. What kind
> of algorithms are used in selecting a specific star (eg. the brightest)
> from a picture in a CCD plate? I heard about Newton's 2D application...?
> Is there any good pages of these topics in Internet?

Don't have a web reference, but there are basically two approaches for
star trackers; you can either try to find a specific (generally very
bright) star or stars, or you can do some type of pattern recognition on
multiple stars.  The former approach was common in days gone by, when
onboard computer power was limited, and frequently used Capella as a
reference star, as I recall; the general idea is that you can always find
the Sun, and that plus some coarse knowledge of your position (e.g.,
knowing the date, which gets you the Sun-Earth line orientation to ~1
degree if you're anywhere in Earth orbit) will let you scan by rotating
the tracker (or the spacecraft) and locking onto the brightest object you

Pattern recognition on multiple stars involves a trade between field of
view (and accuracy) and sensitivity, with additional factors for computer
power, available time, initial conditions, etc.  Most imaging star
trackers have fairly small fields of view (a few degrees) and must be
quite sensitive to ensure that at least two stars are visible no matter
where they're looking; in some cases they can be "lost" if they happen to
look at an empty part of the sky.  But they're quite accurate (a few
microradians) once they lock onto a star.  Given the large number of stars
they need to be able to lock on to, they require a large star catalog and
may need some prior information (e.g. from a sun sensor and Earth sensor)
or a long time and some scanning around to find a configuration they
recognize.  To my knowledge, they all use the brightest stars in the
field, but they may or may not use the brightness information as part of
the algorithm for recognizing star fields.

The Clementine/Brilliant Pebbles star trackers were required to be able to
determine orientation very quickly (100 ms) without any prior knowledge
(the "lost in space" problem).  They used a very wide-field lens (It's
been a while, but I think it was 40 x 24 degrees) so that they only needed
a catalog of a few hundred stars to ensure that at least three were in the
field of view.

The actual attitude finding algorithm for the Clementine tracker selected
the 5 brightest starlike objects in the field and defined triangles
between sets of three.  It did the pattern matching only on the basis of
the ratio of sides of the triangles, without using brightness information
or absolute angle information, which made the algorithm both fast and very
robust -- but seemed to baffle some of the JPL folks who looked at it,
since they kept wanting to know about the linearity of the sensor

Jordin Kare

From: (Henry Spencer)
Subject: Re: On-board instrument pointing systems?
Date: Mon, 17 Jul 2000 19:04:09 GMT

In article <>,
Jordin Kare <> wrote:
>...The former approach was common in days gone by, when
>onboard computer power was limited, and frequently used Capella as a
>reference star, as I recall...

Canopus was the usual choice.  It's the second-brightest star in the sky,
well away from the ecliptic (i.e. always at a considerable angle to the
Sun direction for near-Earth and most planetary craft), set in an
otherwise-dim neighborhood that makes finding it and tracking it fairly

Canopus was adopted after negative experiences with Earth sensors on the
Rangers and the first Mariners.  Earth was an obvious target because,
after all, that's where you want to point the antenna... but its
brightness varies a lot with angle and distance, it moves, there are often
other bright objects in the same part of the sky, and it's sometimes not
far from the Sun line.
Microsoft shouldn't be broken up.       |  Henry Spencer
It should be shut down.  -- Phil Agre   |      (aka

From: (Jordin Kare)
Subject: Re: On-board instrument pointing systems?
Date: Sun, 23 Jul 2000 06:47:55 GMT

In article <>, (Henry Spencer)

> In article <>,
> Jordin Kare <> wrote:
> >...The former approach was common in days gone by, when
> >onboard computer power was limited, and frequently used Capella as a
> >reference star, as I recall...
> Canopus was the usual choice.

Thank you, Henry.  I realized I'd remembered wrong about a day after I posted...


Index Home About Blog