(Preprint) AAS 19-031

AN AUTONOMOUS PASSIVE NAVIGATION METHOD FOR

NANOSATELLITE EXPLORATION OF THE ASTEROID BELT

Leonard Vance,* Jekan Thangavelautham,† and Erik Asphaug‡

There are more than 750,000 asteroids identified in the main belt. These aster-
oids are diverse in composition and size. Some of these asteroids can be traced
back to the early solar system and can provide insight into the origins of the so-
lar system, origins of Earth and origins of life. Apart from being important tar-
gets for science exploration, asteroids are strategically placed due to their low-
gravity well, making it low-cost to transport material onto and way from them.
They hold valuable resources such as water, carbon, metals including iron, nick-
el and platinum to name a few. These resources maybe used in refueling depots
for interplanetary spacecraft and construction material for future space colonies,
communication relays and space telescopes. The costs of getting to the main as-
teroid belt, combined with large numbers of objects to be explored encourage
the application of small spacecraft swarms. The size and capability of the result-
ing nano-spacecraft can make detection from Earth difficult. This paper dis-
cusses a method by which a spacecraft can establish ephemeris autonomously
using line of sight measurements to nearby asteroids with Extended Kalman Fil-
tering techniques, without knowing accurate ephemeris of either the asteroids or
the spacecraft initially. A description of the filter implementation is followed by
examples of resultant performance.

INTRODUCTION
The number of asteroids currently cataloged exceeds seven-hundred fifty thousand. These as-
teroids are known to be diverse in size, composition and origin. Asteroids surfaces undergo
weathering and are covered by dust that hide their internal composition. Older asteroids are
thought to undergo the ‘Brazil Nut’ effect, where large boulders settle to the top, leaving a differ-
ent composition of material below. Ground observation and even in-space observation cannot be
used to guarantee what is beneath these top layers. Systematically performing close-up surveys,
sample analysis/return and use of bistatic radar is perhaps the only sure-way of determining the
internal composition of these asteroids. The daunting task of exploring any statistically signifi-
cant subset of this total will likely involve swarms of small spacecraft, which would place a chal-
lenging requirement on any terrestrial based navigation scheme implemented. A technique for
some type of autonomous navigation would be useful, allowing spacecraft to navigate within the
asteroid belt without support from an Earth based system.

* PhD Candidate, Aerospace and Mechanical Engineering, Univ. of Arizona, 1130 N Mountain, Tucson, AZ 85721
† Assistant. Professor, Aerospace and Mechanical Engineering, Univ. of Arizona, 1130 N Mountain, Tucson, AZ 85721
‡‡ Professor. Lunar and Planetary Laboratory, Univ. of Arizona, 1629 E University, Tucson AZ 85721

1
 Navigation techniques using passive line-of-sight measurements are well known within the
community. The ability to infer ephemeris information of an orbital object by tracking it from a
known position is well understood, as is the ability to infer own ephemeris by tracking an object
with known position and velocity. This paper shows that it is also possible, however, to infer
both own-ship and tracked-object ephemeris given knowledge of the gravitational influences both
are subjected to.
Figure 1 shows the general nature of the problem. It is well established that position and ve-
locity data for an overhead satellite can be extracted from passive line of sight measurements
from the ground. Observability of range is inferred from the gravitational attraction of the orbit-
ing object. Likewise, as a simple extension, it is also possible to do the same job from an orbiting
satellite. If you know your own position and velocity, you can infer the position and velocity of
another satellite given a time history line of sight measurements.

Method 1: Position and Velocity Method 2: Position and Velocity Method 3: (Subject of this Paper)
inferred from line of sight inferred from line of sight Position and Velocity of both
measurements from ground measurements from satellite with satellites inferred from line of
station with known position known position and velocity sight history

Figure 1: Auto-navigation using other objects under gravitational influence is the logical
descendent of established navigation techniques

With some additional thought, it is possible that the position and velocity of both the target
and ownship satellites can be inferred from a single line of sight history. With some exceptions,
it can be asserted that any line of sight history between two objects gravitationally influenced by a
common source is the result of only one specific target and home satellite trajectory. Further-
more, the tracking of more than one object improves navigational performance, and this paper
evaluates the utility of tracking a 2nd object (an asteroid in this case).

RELATED WORKS
Current missions beyond the Earth-Moon system utilize the Deep Space Network or equivalent
network of ground stations to determine position in space and perform navigation through Dop-
pler ranging. As more small-space missions are planned and many active spacecraft in operation,
the DSN will be under strain and there will be an important need to find viable alternatives. A
substantial body of work exists regarding deep space auto-navigation. The most popular method-
ology is triangulation or inferential estimation by recording the line of sight history of a known
object from the spacecraft. Others propose the precise measurement of pulsar time histories, and
some propose range measurements to crosslink spacecraft or optical measurement of known ce-
lestial objects combined with doppler measurements of solar output. None of these techniques
propose utilizing passive line of sight angular tracking to objects with poor position knowledge.

2
 Multiple references are available which explore the ability to infer ownship position and ve-
locity via line of sight history of an object with well known location. S. Bhaskaran et al.4,
Karimi et al.7, Polle et al.8, and Riedel et al9 all propose some form or combination of this tech-
nique, tracking one or more objects with known position. K. Hill and G. H. Born5 propose range
measurements to cross link vehicles with known position, while Yim et al.3 combines angular
measurements to the earth and spectrometer measurements of doppler shift from the sun. Finally,
a body of work represented by Shemar et al.10 explore the possibility of using x-ray pulsars,
measuring the arrival of pulses and thus inferring distance from them. This approach would then
work much like GPS, triangulating using the change in linear distances to 3 or more neutron stars
to establish position.

METHOD
The Extended Kalman Filter (EKF) is a well-known, pseudo-optimal predictor-corrector filter
widely used for a wide variety of aerospace purposes. This paper shows the adaptation of the
basic filter structure for this particular problem. The established form of the EKF* starts by prop-
agating the current estimate and covariance matrices forward by one timestep.

𝒙𝑘 1𝑘 𝒙𝑘𝑘 𝒇 𝒙 𝑡 𝑡 , 𝒖∗ 𝑡 ,𝑡 𝑑𝑡 (1)

𝑷𝑘 1𝑘 𝜱 𝑘 1, 𝑘 𝑷 𝑘 𝑘 𝜱 𝑘 1, 𝑘 𝑸 𝑘 (2)

This followed by calculation of the Kalman gains:

𝑲 𝑘 1 𝑷 𝑘 1𝑘 𝑯 𝑘 1𝑷𝑘 1𝑘 𝑯 𝑘 1 𝑹 𝑘 1 (3)
The state vector estimate can then be updated:
𝒙 𝑘 1𝑘 1 𝒙 𝑘 1𝑘 𝑲 𝑘 1 𝒛 𝑘 1
(4)
𝒉 𝒙 𝑘 1 𝑘 , 𝒖∗ 𝑘 1 , 𝑘 1 𝑯𝒖 𝑘 1 𝜕𝒖 𝑘 1
Finally, also update the covariance matrix:
𝑷𝑘 1𝑘 1 𝑰 𝑲 𝑘 1𝑯 𝑘 1 𝑷 𝑘 1|𝑘 (5)

The definition of these parameters are consistent with normal EKF usage:
𝒙 𝑠𝑡𝑎𝑡𝑒 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒
k current time step
k 1 next time step
f function propagating state estimate 𝒙 in time
t time
u* nominal input
P Covariance Matrix
Φ Linearized time propagation matrix
𝑸𝒅 𝑆𝑡𝑎𝑡𝑒 𝑒𝑟𝑟𝑜𝑟 𝑝𝑟𝑜𝑝𝑎𝑔𝑎𝑡𝑖𝑜𝑛 𝑚𝑎𝑡𝑟𝑖𝑥
K Kalman gain matrix
h function taking state vector to measurements
𝑯𝒙 Jacobean of h w/r to state vector x
𝑯𝒖 Jacobean of h w/r to external inputs
I Identity matrix

* “Lessons in Digital Estimation theory”, Jerry M. Mendel, Prentice-Hall, 1987

3
 The function u* and its derivatives are zero for the purposes of this exercise since there are no
external forces acting on the system, and as such, (1) through (5) can be simplified to give

𝒙 𝑘 1𝑘 𝒙𝑘𝑘 𝒇 𝒙 𝑡 𝑡 , 𝑡 𝑑𝑡 (6)

𝑷𝑘 1𝑘 𝜱 𝑘 1, 𝑘 𝑷 𝑘 𝑘 𝜱 𝑘 1, 𝑘 𝑸 𝑘 (7)

for the propagation equations, and

𝑲 𝑘 1 𝑷 𝑘 1𝑘 𝑯 𝑘 1𝑷𝑘 1𝑘 𝑯 𝑘 1 𝑹 𝑘 1 (8)
𝒙 𝑘 1𝑘 1 𝒙 𝑘 1𝑘 𝑲 𝑘 1 𝒛 𝑘 1 𝒉𝒙𝑘 1𝑘 } (9)
𝑷𝑘 1𝑘 1 𝑰 𝑲 𝑘 1𝑯 𝑘 1 𝑷 𝑘 1|𝑘 (10)

for the update equations.

The state estimation vector 𝒙 contains 3 element position, velocity and accelerations for each
object. Since this paper discusses tracking two objects from a home vehicle, there are 27 ele-
ments in the state estimation vector comprising:
(3x1)
(9x1) 𝑥 𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒, 𝑡𝑟𝑎𝑐𝑘𝑖𝑛𝑔 𝑣𝑒ℎ𝑖𝑐𝑙𝑒
each 𝒓𝑻 𝑦 𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒, 𝑡𝑟𝑎𝑐𝑘𝑖𝑛𝑔 𝑣𝑒ℎ𝑖𝑐𝑙𝑒
𝑧 𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒, 𝑡𝑟𝑎𝑐𝑘𝑖𝑛𝑔 𝑣𝑒ℎ𝑖𝑐𝑙𝑒
𝒙𝑻 𝒗𝑻 x,y,z velocity estimates, 𝑡𝑟𝑎𝑐𝑘𝑖𝑛𝑔 𝑣𝑒ℎ𝑖𝑐𝑙𝑒
(27x1)

𝒙 = 𝒂𝑻 x,y,z acceleration estimates, 𝑡𝑟𝑎𝑐𝑘𝑖𝑛𝑔 𝑣𝑒ℎ𝑖𝑐𝑙𝑒

𝒙𝟏 Object 1 estimates
𝒙𝟐 Object 2 estimates

The function f is a simple orbital propagator, utilizing the basic gravitational law
𝐺𝑚 𝒓
𝒂 (11)
𝒓
where the r’s are taken from the position estimates in the state vector. The corresponding matrix
𝜱 is the Jacobean of this with respect to 𝒙 , giving a 27 × 27 matrix. The function h takes state
variables and constructs the measurements z, and is therefore of the form:
𝒛 𝒉 𝒙 (12)

Since the measurements for this system are the normalized line of sight vectors from the tracking
vehicle to the two asteroids, the resulting function h is

4
 (13)

The Jacobian of this function with respect to each component of the state estimate vector 𝒙 pro-
vides the resulting 6 × 27 Hx matrix.
The R matrix represents the system measurement noise, and this takes the form of angular uncer-
tainties along the line of sight to objects 1 and 2. The measurement vector z is the two normal-
ized line of sight vectors to each target as given by (13), so given an angular uncertainty ε in each
axis, the covariance matrix for measurement errors to object 1 in line of sight coordinates is:
1 0 0
𝑹𝟏,𝒍𝒐𝒔 0 𝜀 0 (12)
0 0 𝜀
This must then be rotated into the heliocentric coordinate frame used by the rest of the filter.
Derivation of the coordinate transform matrix is done by establishing sequential coordinate frame
rotation angles about the z and then y-axes respectively, providing a coordinate transformation
matrix of
𝑐𝑜𝑠𝜓 𝑐𝑜𝑠𝜃 𝑠𝑖𝑛𝜓 𝑐𝑜𝑠𝜃 𝑠𝑖𝑛𝜃
𝑇, 𝑠𝑖𝑛𝜓 𝑐𝑜𝑠𝜓 0 (13)
𝑐𝑜𝑠𝜓 𝑠𝑖𝑛𝜃 𝑠𝑖𝑛𝜓 𝑠𝑖𝑛𝜃 𝑐𝑜𝑠𝜃

Where the sequential angles of rotation about z and y axes respectively are:
𝑦 𝑦
𝜓 𝑡𝑎𝑛 (14)
𝑥 𝑥
𝑧 𝑧
𝜃 𝑡𝑎𝑛 (15)
𝑥 𝑥 𝑦 𝑦

This matrix takes a vector from inertial to the line of sight coordinated frame with respect to ob-
ject 1. Using a similarity transformation to convert from line of sight to inertial coordinates, we
have
𝑹𝟏,𝒊𝒏𝒆𝒓𝒕𝒊𝒂𝒍 𝑻𝑰,𝒍𝒐𝒔𝟏 𝑹𝟏,𝒍𝒐𝒔 𝑻𝑰,𝒍𝒐𝒔𝟏 𝑻 (16)

The second R matrix to the 2nd object can be similarly calculated, and the two can be combined to
provide the overall 6 × 6 input to the filter equations.

5
 𝑹𝟏,𝒊𝒏𝒆𝒓𝒕𝒊𝒂𝒍 𝟎 (17)
𝑹
𝟎 𝑹𝟐,𝒊𝒏𝒆𝒓𝒕𝒊𝒂𝒍

Finally, the 27 × 27 state process noise matrix Q is constructed with arbitrarily small diagonal
elements for the purpose of establishing feasibility.

𝑸 1.0𝑥10 ∗𝑰 (18)

RESULTS
The implementation of the above filter is tested in an environment derived from the JPL aster-
oid ephemeris database*. The tracking vehicle is established in a circular orbit at the median he-
liocentric altitude, and a collection of nearby asteroids is established to provide a pool of possible
objects to track and test filter performance. The test case chosen utilizes asteroids with catalog
numbers 32,675 and 272,344, and an argument of latitude of π/3 for the tracking spacecraft. The
resulting geometry is shown in Figure 2, along with a history of the distances between the track-
ing spacecraft and the two asteroids. The simulation is then executed for 1 year, sampling once
per hour. The standard deviation of the line of sight error is 10 microradians in vertical and hori-
zontal axes. Initial position and velocity variances are set to (100,000km)2 and (1 km/s)2 respec-
tively.

Asteroid 2

Tracking
Mars Orbit Spacecraft

Earth Orbit Asteroid 1

Figure 2: Test Geometry Overview

Case 1: Single object tracking

This case uses only tracking information from asteroid 1 to establish position and velocity. Re-
sults are shown in Figure 3, illustrating good convergence in position and velocity in all 3 axes.
The associated position and velocity errors for the asteroid tracking are then summarized similar-
ly in Figure 4. Results are similar to those seen for the observer spacecraft, plausibly because the
problem is nearly symmetric given the identical starting covariances.
An overall assessment of filter performance assuming the results actually converge can be ex-
tracted from the time history of covariance magnitudes as shown in Figure 5. The convergence of
the observer and asteroid are nearly identical given the initial condition stated, and steady im-

*
NASA/JPL website for asteroid orbital ephemeris. https://ssd.jpl.nasa.gov/?sb_elem

6
provement in position and velocity knowledge continue well after the closest point of approach of
the tracked asteroid.

Figure 3: Case 1 (single asteroid track) residual error and 3-sigma covariance bounds for
observer position and velocity errors.

Figure 4: Case 1 (single asteroid track) residual error and 3-sigma covariance bounds for
asteroid position and velocity errors.
3 Sigma Error Estimates
1010

Observer Pos
105
Asteroid Pos
Observer Vel
Asteroid Vel

100
0 100 200 300 400
Time (days)
Figure 5: Case 1 error bound convergence

7
Case 2: Two Asteroids Tracked
Case 2 uses the same first asteroid as Case 1 but adds a 2nd asteroid as shown in Figure 2. Once
again, all objects are given the same initial covariances outlined in the previous section and the
simulation is then executed over a 1-year epoch with both asteroids being used to update the
EKF. Basic results for the observer spacecraft are shown in Figure 6, and these show a signifi-
cant improvement in performance over the single asteroid case shown in Figure 3, with noticeable
reductions in the peak errors, and earlier convergence seen in each axis. Overall results for both
the satellite and the two asteroids are given in Figure 7, noting that overall performance improves
by a factor of 2 to 5.

Figure 6: Case 2 (2 asteroids tracked) residual error and 3-sigma covariance bounds for
observer are significantly improved from single asteroid

Figure 7: Case 2 error bounds converge faster

because two asteroids are being tracked

CONCLUSIONS
The concept of achieving auto-navigation using passive line of sight measurements from poor-
ly known objects under the influence of a common gravitational field is demonstrated and quanti-
fied, with results showing convergence for both the observer and asteroid position and velocity
estimates. Significant improvements in performance are observed when the number of tracked
asteroids is increased from one to two, and it is expected that the addition of more observations

8
using other asteroids would improve observability and therefore filter performance even further.
The relatively low tracking rate required for this technique allows the possibility of perhaps doz-
ens of asteroids being tracked, with the expectation of significantly improved accuracy.
Mission design for future swarm asteroid exploration missions can incorporate this auto-
navigation technique to offload Earth-based navigation assets such as the DSN by providing an
inexpensive alternative. In addition, the ephemeris of the tracked asteroids is improved almost as
much as that of the observer, and this information can be used to update the orbital parameters of
every asteroid tracked.

REFERENCES
1 J.L. Doe and J.Q. Public, "The Parameterization of the Rotation Matrix using Redundant Attitude Coordinates." Non-
linear Dynamics. Vol. 32, No. 3, 2005, pp. 71–92.
1 J. M. Mendel, "Lessons in Digital Estimation Theory”, Prentice-Hall Press, 1987, pp. 252-253.
2 NASA/JPL website for asteroid orbital ephemeris. https://ssd.jpl.nasa.gov/?sb_elem
3J. R. Yim, J. L. Crassidis and J. L Junkins, "Autonomous Orbit Navigation of Interplanetary Spacecraft”, AIAA-2000-
3936.
4 S. Bhaskaran, J.E. Riedcl, S. P. Synnott, "Autonomous Optical Navigation for Interplanetary Missions" SPIE Pro-

ceedings. 1996.
5K. Hill and George H. Born, " Autonomous Interplanetary Orbit Determination using Satellite to Satellite Tracking."
Journal of Guidance, Control, and Dynamics. Vol. 30, No. 3, May-June 2007.
6 B.Polle, B.Frapard, X.Sembely, "Autonomous Navigation Concepts for Interplanetary Missions”, IFAC Proceedings

Volumes. Vol. 37, Issue 6, June 2004, pp. 203-208.

7 R. R. Karimi and D. Mortari, "Interplanetary Autonomous Navigation Using Visible Planets" Journal of Guidance,

Control, and Dynamics, Vol. 38, Issue 6, 2015, pp. 1151-1156.

8 B. Polle, B. Frapard, J. Gil-Fernandez, E. Milic, M Graziano, J. Rebordao, P. Motrena, " Autonomous Navigation for
Interplanetary Missions, Performance Achievements Based on Real and Flight Images " Proceedings of the 6th interna-
tional ESA Conference on Guidance, Navigation and Control Systems. 17-20 October 2005.
9J. E. Riedel, S. Bhaskaran, S. Desai, D. Han, B. Kennedy, T. McElrath, G. W. Null, M. Ryne, S. P. Synnott, T. C.
Wang, R. A. Werner, "Using Autonomous Navigation for Interplanetary Missions: The Validation of Deep Space I
Autonav." International Conference on Low-Cost Planetary Missions. May 2, 2000.
10S. Shemar et al, " Towards practical autonomous deep-space navigation using X-Ray pulsar timing " Experimental
Astronomy. Vol. 42, No. 2, October 2016, pp.101-138.