⚡ electrical engineering

Star-Tracker-Constrained Attitude MPC for CubeSats

This paper proposes an online linear model predictive control framework that analytically linearizes nonlinear spacecraft dynamics and geometric constraints to efficiently perform slew maneuvers while maintaining star-tracker availability during ground-target tracking for CubeSats.

Original authors: Dominik Beňo, Patrik Valábek, Martin Hromčík, Martin Klaučo

Published 2026-04-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Dominik Beňo, Patrik Valábek, Martin Hromčík, Martin Klaučo

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are holding a high-powered camera on a spinning top, trying to take a perfect photo of a specific house on the ground while the top is moving fast. But there's a catch: you have a very sensitive eye (a star-tracker) attached to the top that helps you know which way is "up."

Here's the problem:

The Eye is Fussy: If you spin the top too fast, the eye gets dizzy and can't see the stars anymore. If you point the eye directly at the Sun or the ground, it gets blinded.
The Camera is Picky: You need to keep the camera perfectly aimed at the house, even as the top moves.
The Brain is Small: This isn't a supercomputer; it's a tiny CubeSat (a mini-satellite the size of a shoebox) with very limited brainpower.

This paper presents a new "brain" (a control system) for these tiny satellites that solves this puzzle in real-time. Here is how it works, broken down into simple concepts:

1. The Old Way: The "Pre-Planned Road Trip"

In the past, if a satellite needed to turn, engineers on Earth would calculate the perfect path days in advance. They would say, "Okay, spin slowly here, stop there, turn fast there," and upload that list of instructions.

The Flaw: If the satellite sees a new target it wasn't expecting, or if the satellite is slightly heavier than they thought, the pre-planned path fails. The satellite can't react quickly because it has to wait for a new list of instructions from Earth.

2. The New Way: The "Smart GPS" (Model Predictive Control)

The authors created a system that acts like a smart GPS navigation app. Instead of following a pre-written list, the satellite constantly asks itself: "Where am I? Where do I need to go? What obstacles are in my way right now?"

It does this using a method called Model Predictive Control (MPC). Think of it like a chess player who looks 5 moves ahead.

The Look-Ahead: Every 0.1 seconds, the satellite simulates the next few seconds of its own movement in its head.
The Rules: It checks three main rules during this simulation:
1. Don't spin too fast (or the star-tracker goes blind).
2. Don't point at the Sun or Earth (or the star-tracker gets blinded).
3. Don't use too much fuel/power (the motors have limits).

3. The Magic Trick: "The Sketch vs. The Painting"

The hardest part of this problem is that space physics is messy and non-linear (like trying to draw a perfect circle freehand). Doing the complex math for every single second would crash the satellite's tiny computer.

The authors' clever solution is Linearization.

The Analogy: Imagine you are driving a car on a winding mountain road. To plan your next turn, you don't need to map out the entire winding road. You just look at the tiny patch of road right in front of you. On that tiny patch, the road looks straight.
The Application: The satellite assumes the physics are "straight" (simple) for just the next few seconds. It solves a simple math problem (a Quadratic Program) to find the best move. Then, a split second later, it updates its position, looks at the new "straight patch," and solves the simple math again.
The Result: By constantly re-calculating simple problems, it navigates the complex, winding reality perfectly without needing a supercomputer.

4. The "Slack" Safety Net

Sometimes, the rules are so tight (e.g., "Don't spin fast AND don't look at the Sun AND hit the target") that a perfect solution doesn't exist.

The system uses "slack variables." Think of these as safety buffers. If the satellite is almost breaking a rule, the system allows a tiny, temporary violation but penalizes it heavily in the math. This ensures the satellite never gets "stuck" or crashes because it couldn't find a perfect answer. It keeps moving forward safely.

5. The Results: Did it Work?

The team tested this in a high-fidelity computer simulation (using NASA's "42" simulator) that mimics real space conditions, including wind, gravity, and imperfect satellite parts.

Comparison: They compared their new "Smart GPS" against the old "Pre-Planned Road Trip" and a "Naive" method (just turning as fast as possible without thinking).
The Winner: The new system was almost as accurate as the complex pre-planned method but could react instantly to changes.
The "Blind Spot" Limitation: They found one tricky scenario: If the target is in a spot where the Sun is on one side and the Earth is on the other, the satellite literally cannot point its camera at the target without blinding its eye. This is a physical law, not a software bug. The system correctly identified this impossible situation rather than crashing.

Summary

This paper gives small, cheap satellites a "smart pilot" that can look ahead, avoid blinding itself, and keep its camera steady on moving targets—all while running on a tiny computer. It turns a complex, messy physics problem into a series of simple, quick decisions, allowing CubeSats to be much more agile and autonomous in space.

1. Problem Statement

The paper addresses the challenge of agile ground-target tracking for small spacecraft (specifically CubeSats) equipped with a single star-tracker.

The Conflict: To maintain high-quality data from optical instruments (e.g., push-broom cameras, SAR), the spacecraft must continuously reorient to follow a moving ground target. However, a single star-tracker has strict operational limits:
1. Rate Limits: It cannot provide reliable attitude data if the spacecraft's angular rate exceeds a specific threshold (typically $1^\circ/s$ to $3^\circ/s$ ).
2. Exclusion Zones: The sensor must avoid pointing at bright extended bodies, specifically the Sun and the Earth's limb (nadir), to prevent blinding.
The Gap: Existing solutions rely on offline trajectory planning (lacking responsiveness to unplanned targets) or nonlinear Model Predictive Control (NMPC), which is computationally expensive and difficult to certify for fixed-cycle onboard software. There is a lack of online feedback controllers that simultaneously enforce pointing accuracy, angular rate limits, actuator constraints, and star-tracker exclusion zones in real-time.

2. Methodology

The authors propose an online, time-varying linear Model Predictive Control (MPC) framework. The core innovation is transforming a complex nonlinear constrained problem into a computationally efficient Quadratic Program (QP) solvable on CubeSat-class hardware.

A. System Modeling & Linearization

Dynamics: The spacecraft is modeled as a rigid body with 8U CubeSat inertia properties. The state vector includes angular velocity ( $\omega$ ) and quaternion ( $q$ ).
Analytical Linearization: Instead of solving a full nonlinear optimization at every step, the authors analytically linearize the nonlinear rigid-body dynamics and geometric constraint outputs (pointing error, Sun/nadir angles) about the current state estimate at each control step.
Result: This yields a Time-Varying Linear MPC formulation. The prediction horizon uses a discrete linear model updated at every sampling instant ( $T_s = 0.1s$ ).

B. Observer Design

A Time-Varying Kalman Filter (TVKF) is employed to estimate the state and augmented disturbance channels.
Disturbance Modeling: The state is augmented with:
- Input disturbances ( $d_u$ ) to model unmodeled torques (e.g., gravity gradient, magnetic).
- Output disturbances ( $d_y$ ) to compensate for linearization errors and geometric biases.
This ensures robustness against model-plant mismatches.

C. Control Formulation

The MPC problem is cast as a standard Quadratic Program (QP):

Objective Function: Minimizes pointing error (deviation of instrument boresight from target), angular rate magnitude, and control effort (torque changes).
Constraints:
- Hard Constraints: Actuator torque limits ( $u_{max}$ ).
- Softened Constraints: Angular rate limits and geometric exclusion zones (Sun/Nadir) are enforced using slack variables with a very high penalty weight ( $w_s = 10^9$ ). This prevents QP infeasibility while strictly prioritizing constraint satisfaction.
Receding Horizon: The controller solves the QP over a prediction horizon ( $N=50$ steps), applies only the first control input, and repeats.

3. Key Contributions

Online Constraint-Aware Tracking: A novel MPC framework that jointly enforces pointing, angular rate, actuator, and star-tracker exclusion constraints in a single online feedback loop, eliminating the need for offline trajectory design.
Computational Efficiency: By using analytical linearization at each step, the complex nonlinear problem is reduced to a standard QP. This offers a significantly lower computational profile than nonlinear MPC (NMPC), making it suitable for low-power CubeSat processors.
Robustness Characterization: A comprehensive Monte Carlo (MC) campaign characterizing the controller's performance under varying target geometries and significant inertia perturbations ( $\pm 30\%$ ).

4. Simulation Results

The controller was validated using NASA's "42" high-fidelity spacecraft dynamics simulator in a Model-in-the-Loop (MIL) configuration.

A. Comparison Scenarios

The proposed Linear MPC was compared against:

Offline Trajectory Optimization: A global nonlinear optimal control solution (computed on the ground).
Naive Solution: A reference-shaping approach that limits rate but ignores exclusion zones.

B. Performance Metrics (Single Run)

Constraint Satisfaction: The Linear MPC successfully maintained the star-tracker within exclusion zones and respected rate limits, whereas the "Naive" solution violated rate limits by ~3% and ignored exclusion zones.
Pointing Accuracy:
- Linear MPC: Mean pointing error of 0.188°.
- Offline Optimization: Mean error of 0.031°.
- Naive: Mean error of 0.035°.
- Note: While MPC had slightly higher error due to linearization approximations, it remained well within the required $1^\circ$ tolerance and provided real-time feedback capabilities the offline method lacked.
Settling Time: MPC settled in ~49.7s, comparable to the other methods.

C. Monte Carlo Analysis (100 Runs)

Constraint Satisfaction: Exclusion constraints were active in 97% of runs and satisfied in 100% of runs.
Geometric Limitation: In 11 runs, the simultaneous activation of Sun and Nadir exclusion zones forced the instrument boresight to drift, causing pointing errors $>1^\circ$ . This is identified as a fundamental geometric limitation of single-star-tracker configurations, not a controller failure.
Robustness: In the remaining 89 valid runs, the mean pointing error was 0.31°, with 65 runs maintaining error $<1^\circ$ throughout the entire observation window.
Computational Load: The QP solver required an average of 19.28 iterations per control step, confirming the method's suitability for real-time implementation.

5. Significance and Conclusion

This paper demonstrates that online, constraint-aware attitude control is feasible for resource-constrained CubeSats without relying on heavy ground intervention.

Operational Autonomy: The method enables spacecraft to react to unplanned targets or changing geometries autonomously, a critical capability for future agile Earth observation missions.
Flight Heritage Compatibility: The formulation as a standard QP aligns with established aerospace flight software practices, facilitating easier certification compared to iterative nonlinear solvers.
Future Work: The authors plan to integrate realistic sensor noise (gyros, star-trackers) and reaction wheel models to validate the observer, as well as develop strategies to handle geometrically infeasible configurations (where exclusion zones make tracking impossible).

In summary, the proposed Star-Tracker-Constrained Attitude MPC successfully bridges the gap between high-fidelity constraint handling and the computational limitations of small satellites, offering a robust solution for continuous, agile ground tracking.