AIAA 2015-4422

SPACE Conferences and Exposition

31 Aug-2 Sep 2015, Pasadena, California
AIAA SPACE 2015 Conference and Exposition

An Intelligent Attitude Determination and Control System

Concept for a CubeSat Class Spacecraft

Jeremy Straub
Department of Computer Science, University of North Dakota, Grand Forks, ND 58202

An attitude determination and control system (ADCS) is used to orient a spacecraft for a
wide variety of purposes (e.g., to keep a camera facing Earth or orient the spacecraft for
propulsion system use). The proposed intelligent ADCS has several key features: first, it can
be used in multiple modes, spanning from passive stabilization of two axes and
unconstrained spin on a third to three-axis full active stabilization. It also includes
Downloaded by CARLETON UNIVERSITY on November 2, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-4422

electromagnetic components to ‘dump’ spin from the reaction wheels. Second, the ADCS
utilizes an incorporated autonomous control algorithm to characterize the effect of actuation
of the system components and, thus, learns its movement model. This allows it to respond to
changing conditions (such as damage, deployable parts, depleted fuel, etc.) that impact the
movement model as well as facilitating the learning of the initial model, based on craft
manufacturing nuances.

I. Introduction

A TTITUDE determination and control is a critical subsystem for small and large spacecraft alike. The
spacecraft’s capability for imaging the Earth, other space-based objects and even many forms of
communications is driven – in part – by its attitude determination and control capabilities. Some forms of
communication and wireless power transfer, for example, require very accurate pointing control due to limited beam
spread (minimizing signal loss over distance). Other spacecraft may require accuracy for cluster operations or other
mission needs.
Despite the benefits possible from highly accurate pointing, many CubeSats lack these capabilities. Some have
unconstrained rotation on one or more axes; others use passive stabilization techniques or a combination of passive
stabilization on some axes with active stabilization on one or more axes. Yet others use attitude determination and
control systems (ADCS) with low granularity sensing and/or pointing levels or more robust systems, both of which
require significant calibration on the ground and prospective refinement on orbit.
An ADCS whose control is based on machine learning principles and iterative refinement is proposed herein.
The system is an active three-axis ADCS control with an accurate IMU and magnetometer for attitude
characterization (aided by an onboard camera that is a payload system component) and three reaction wheels and
magnetorquers for attitude control. The system is able to run in a variety of active, partially active and simulated
passive control modes.
Movement model determination is performed onboard the spacecraft via making and characterizing small
actuations of the reaction wheels and magnetorquers and characterizing the movement results. A basic movement
model is then augmented and continuously refined based upon the information collected. This refinement capability
allows automatic adaptation to a changing spacecraft configuration, due to propellant (or other) depletion,
deployment or unexpected occurrences. Incorporating movement model refinement onboard the spacecraft also
mitigates the risk of an occurrence creating a spin, etc. that impairs the communication required to obtain an updated
ground-calculated model. This paper presents the proposed ADCS as well as high-level algorithms for its control.

II. Benefits and Background

While the use of any autonomy on a spacecraft may not be accepted without some concern [1, 2], autonomy has
found numerous uses on numerous prior missions [3-5]. Autonomy control algorithms have found use in fault
diagnosis [6], rendezvous [7], interplanetary trajectory optimization [8] and numerous other applications. Specific
to the automation of attitude determination and control, several prior studies have been conducted. Cheng [9] and
Van Buijtenen [10] demonstrated the use of fuzzy logic controllers. Hu, et al. [11] performed work on increasing
fault tolerance with autonomy, as did Lee and Kim [12], and Guan, Liu and Liu [13].

1
American Institute of Aeronautics and Astronautics

Copyright © 2015 by Jeremy Straub. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
 Despite these examples, many ADCS systems do not incorporate autonomy. Movement models for non-
adaptive ADCS units must be pre-programmed, tested and potentially manually adapted if something changes.
Unlike the static (or manually adjusted) approach, the learning approach allows great flexibility in deployment
including (1) not having to perform numerous calculations to develop a movement model, (2) not having to re-
characterize the movement model if a change (expected or unexpected occurs), (3) onboard response capability to
change that may impact ability to point and communicate with the Earth (which would prevent a spacecraft loss
scenario based on something happening and needing Earth controller communications to correct the movement
model). A system implementing this approach is described in the following sections.

III. System Description

The intelligent attitude determination and control system is based on the application of artificial / computational
intelligence techniques to the problem of attitude determination and control. The initial implementation of this
system uses an expert systems framework (see, e.g., [14]); however, other approaches (such as those using fuzzy
logic, etc.) are also possible.
Downloaded by CARLETON UNIVERSITY on November 2, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-4422

A. Basic Operations
The system utilizes a basic movement model, which it initially seeks to validate / update and which is then
updated as normal operations continue. This movement model, combined with an expert systems rule-fact network,
is used to determine how to effect various movement goals. A movement goal is presented as a desired end
orientation. This is de-composed by the movement command processing system into a difference between the
current and target (goal) orientations. This difference is then utilized to determine multiple orientation change
solutions (different combinations of actuator activations) that could be used to achieve this goal. These solutions are
then evaluated based on factors including cost, risk, end state desirability and (if complete satisfaction is not possible
for some solutions) completeness.
The prospective solutions will differ both in the paths that are taken and the actuators that are used to create the
change. Figure 1 illustrates this concept showing that multiple paths can be taken, with regards to rotation in a
single dimension – reality is, of course, more complex as there are three dimensions.

Figure 1. Different Ways of Getting to Target Orientation.

2
American Institute of Aeronautics and Astronautics
 In systems with multiple actuators of a single type, there may be multiple actuation combinations that produce
the same path. Some systems, additionally, may employ multiple ADCS actuation technologies. For example, for a
given movement path, multiple combinations of reaction wheels and magnetorquers could be used to effect the
change. While producing the same movement path, the choice of different actuation approaches may have different
power (or other resource) requirements and create a more or less desirable end state. Using reaction wheels to create
spin and the magnetorquers to dampen it may have the effect of requiring more or less ongoing motion from the
reaction wheels (and have a significant impact on long-term power needs). The impact of actuator combinations
may be different given the movement path chose and current state of the spacecraft. For example, a reaction wheel
(first) / magnetorquer (second) approach may be beneficial for movement in one direction and have the opposite
effect for a movement path in the opposite direction. A diagram of the components of a system to implement this
functionality is presented in Figure 2. Note that, in addition to basic components (such as the processor, memory, a
communications device and storage) it includes sensors and actuators as well as several processing components.
The movement goal command receiving mechanism accepts a user’s (or other subsystem’s) movement commands.
The movement model storage system stores the current movement model (which may have changed significantly
Downloaded by CARLETON UNIVERSITY on November 2, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-4422

from the initially supplied one). The movement determination processing system is used to assess the impact of
various actuator’s actuation on spacecraft orientation. Finally the movement command processing system actually
effects the changes commanded by the user, based on the current movement model.

Processor Actuator Device(s)

Instructions

Memory Sensor Device(s)

Instructions

Movement
Non-Volatile Storage Determination
Bus

Instructions Processing System

Movement Goal
Movement Command
Command Receiving
Processing System
Mechanism

Movement Model
Communications Device
Storage System

Figure 2. System Diagram.

B. Learning How to Operate

If the movement model for the spacecraft was known a priori and did not change, then an expert system (or other
control system) could be developed prior to flight that could contain a set of rules and facts to enable the maneuver
decision making process. In each case, the goal and current state would be presented to the system and a solution
would be provided. This system would embody the movement model and also provide movement command
processing system functionality.

3
American Institute of Aeronautics and Astronautics
 The reality of operations of most spacecraft, however, complexities things significantly. Several factors must be
considered. First, resources on the spacecraft (e.g., propellant) may be consumed over time. This may change the
center of mass of the spacecraft as well as its overall mass and, thus, change the impact of actuation. Second, the
effectiveness of actuators may change (e.g., diminish) over time; thus, sending the same command at the start of the
mission and at the end of the mission may have different effects. Third, actuators may break or suffer significant
impairment. This last category of change may render a subset of prospective movement solutions impossible or
increase their risk (e.g., of non-completion or further damage to the actuator) significantly. A more complex system
is needed to deal with these complexities.
To this end, a correction mechanism for the expert system (the algorithm for which is shown in Figure 3) will be
included that will change the rule network based on changing conditions. For the initial implementation, this
mechanism is a second expert system that evaluates the efficacy of the decision making of the first (though other
technologies could prospectively be used for this component as well). It is be threshold-based and, thus, will only
make changes to the primary expert system when a single error surpasses a mission-defined acceptable level (or a
low level of error persists for a mission-defined length of time). By changing primary expert system rules, the
Downloaded by CARLETON UNIVERSITY on November 2, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-4422

movement model for the spacecraft will be changed allowing future movement goal solutions to be found that more
directly arrive at the desired end state (as opposed to having to make an inaccurate maneuver and then take
corrective action to fix over/under movement).

Start

Determine Model
Learning Command

Send Command to
Spacecraft

Assess Movement
Yes
Caused by Command

Update Movement
Model

Did Movement Model

Change Significantly

No

End

Figure 3. High Level Algorithm.

4
American Institute of Aeronautics and Astronautics
 The speed of change of the movement model must be controlled to prevent sensor failures and temporary
actuator malfunctions from dramatically altering the movement model. Alternately, the movement model must be
able to be changed quickly enough to facilitate it keeping pace with actual system changes (considering the
frequency of ADCS use and model validation and correction frequency). To this end mission-specific weighting
thresholds are used to constrain the level of impact applied to the existing movement model and the difference
between the expected and actual movement when updating the movement model. This is shown in Figure 4.

Existing
Movement
Model

Movement
Command
Downloaded by CARLETON UNIVERSITY on November 2, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-4422

Expected Actual
Movement Movement
Existing
Movement
Difference
Model

Weighting Weighting
Factor Factor

New
Movement
Model
Figure 4. Data Flow.

The complete system combines multiple sensors, actuators and data storage and processing components with logic
that determines when changes are significant enough to merit a change to the movement model being made. A high
level overview of the operation of the whole system is shown in Figure 5.

Magnetometer
IMU Sensor GPS Sensor
Sensor

Movement
Result

Reaction
Wheel(s)
Movement Movement Significantly
Command Expectation Different
Magnetorquer(s)

Yes

Determine
Difference

Apply
Portion of
Difference
Movement
Model

Figure 5. System Component Interactions.

5
American Institute of Aeronautics and Astronautics
 IV. Conclusions and Future Work
This paper has presented a concept for an intelligent attitude determination and control system based on artificial
intelligence (and, prospectively, computational intelligence) techniques. It has described how the system would
operate and discussed several considerations relevant to its implementation. This system offers several forms of
benefit. First, it reduces the reliance on the a priori movement model. This makes the spacecraft resilient to damage
that may occur during launch or deployment (or deployment failures). It also incorporates a de facto check out
mechanism into the system: the characterization of performance before the initial substantive use. Additionally, the
reduced reliance on the a priori movement model may allow the level of testing on the ground to develop and
validate this model to be reduced.
Second, the proposed approach facilitates responsiveness (on a configurable level of speed) to changes in the
spacecraft and its operations. These include both expected changes (such as reduced mass and a shifted center of
mass from propellant depletion) and unexpected ones (such as actuator, sensor and other failures). This increases
the prospective longevity of the spacecraft and may allow the spacecraft to restore ground communications in
scenarios where a failure of a sensor or actuator (onboard a spacecraft that has a movement model that requires
Downloaded by CARLETON UNIVERSITY on November 2, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-4422

controller intervention to update) would otherwise leave the directional communications system pointing in the
wrong direction.
Work is ongoing to develop both software and hardware-software prototypes of this system for terrestrial testing.
The eventual testing in orbit, onboard a small spacecraft is planned.

Acknowledgments
Small satellite development work at the University of North Dakota is or has been supported by the North
Dakota Space Grant Consortium, North Dakota NASA EPSCoR, the University of North Dakota Faculty Research
Seed Money Committee, North Dakota EPSCoR (NSF Grant # EPS-814442), the Department of Computer Science,
the John D. Odegard School of Aerospace Sciences and the National Aeronautics and Space Administration. The
involvement of the numerous students and faculty from multiple disciplines in this project is gratefully
acknowledged. Thanks is also given to Matthew Russell who identified several additional references which are
discussed in Section 2.

References
[1] J. Straub. Attitudes towards autonomous data collection and analysis in the planetary science community.
Galaxies 1(1), pp. 44-64. 2013.
[2] J. Straub. Analysis of the acceptance of autonomous planetary science data collection by field of inquiry.
Advances in Space Research 55(11), pp. 2708-2718. 2015. DOI: http://dx.doi.org/10.1016/j.asr.2015.02.016.
[3] Paolo/Harland Ulivi (David M.). Robotic Exploration of the Solar System: The Golden Age 1957-1982 2007.
[4] Paolo/Harland Ulivi (David M.). Robotic Exploration of the Solar System: Hiatus and Renewal, 1983-1996
2009.
[5] P. Ulivi and D. M. Harland. Robotic Exploration of the Solar System: Part 3: Wos and Woes, 1997-2003 2012.
[6] L. Friedman. Research at jet propulsion laboratory. AI Magazine 4(4), pp. 58. 1983.
[7] E. M. Hinman and D. M. Bushman. Soviet automated rendezvous and docking system overview. Presented at
Automated Rendezvous and Capture Review. Executive Summary. 1991.
[8] F. A. Zotes and M. S. Peñas. Particle swarm optimisation of interplanetary trajectories from earth to jupiter and
saturn. Eng Appl Artif Intell 25(1), pp. 189-199. 2012.
[9] C. Cheng, S. Shu and P. Cheng. Attitude control of a satellite using fuzzy controllers. Expert Syst. Appl. 36(3),
pp. 6613-6620. 2009.
[10] W. M. Van Buijtenen, G. Schram, R. Babuska and H. B. Verbruggen. Adaptive fuzzy control of satellite
attitude by reinforcement learning. Fuzzy Systems, IEEE Transactions On 6(2), pp. 185-194. 1998.
[11] Q. Hu, B. Xiao and M. Friswell. Robust fault-tolerant control for spacecraft attitude stabilisation subject to
input saturation. Control Theory & Applications, IET 5(2), pp. 271-282. 2011.
[12] H. Lee and Y. Kim. Fault-tolerant control scheme for satellite attitude control system. Control Theory &
Applications, IET 4(8), pp. 1436-1450. 2010.
[13] P. Guan, X. Liu and J. Liu. Adaptive fuzzy sliding mode control for flexible satellite. Eng Appl Artif Intell
18(4), pp. 451-459. 2005.
[14] D. Waterman. A Guide to Expert Systems 1986.

6
American Institute of Aeronautics and Astronautics