Tens of thousands of satellites are expected in orbit by the end of the decade, and most of them still wait for engineers on the ground to make every decision. Operators typically add people as they add spacecraft, a model that becomes harder to sustain as fleets grow and missions become more complex.

Some satellite operators may have only eight minutes of communication with a spacecraft every 45 to 90 minutes. During the remaining time, the spacecraft must continue its mission and respond to unexpected conditions on its own.

The need for autonomy will grow as space missions take on work such as inspection, refueling, maintenance, and delivery. Communication links may be jammed or unavailable, GPS may not work, and engineers cannot retrieve a spacecraft to adjust its guidance, navigation, and control (GNC) system after launch.

CUS-GNC is addressing both challenges with SpacePilot, reusable onboard software that combines GNC with reasoning. Because SpacePilot understands the mission objective, the spacecraft makes its own decisions onboard, responds to changing conditions, and keeps working toward its goal without waiting for the next ground contact. Using AI and novel machine learning algorithms, satellites can complete complex missions autonomously, so a fleet can grow without growing the team that controls it. Simone Chesi, founder of CUS-GNC, describes it as “a taxi driver for space.” The software runs onboard the spacecraft and “flies the spacecraft around the solar system.”

Chesi recognized the opportunity through experience in the space industry. “I saw the same work duplicated among a lot of different companies,” he says. Rather than requiring companies to develop a new GNC software stack for each spacecraft, CUS-GNC provides a reusable foundation for different missions.

Developing Autonomy for Changing Conditions

SpacePilot allows customers to build on guidance, navigation, and control capabilities that CUS-GNC has already developed and tested.

To do that, SpacePilot combines established GNC algorithms with trained neural-network models. One application focuses on fault tolerance. Bianchi describes scenarios in which a spacecraft loses part of its control capability after an equipment failure. The team trains neural-network models to determine how the spacecraft can continue operating with its remaining resources.

Before incorporating the models into SpacePilot, CUS-GNC evaluates them across thousands of simulated operating conditions using Monte Carlo simulations. These tests help the team assess how reliably the models respond to different mission scenarios.

SpacePilot is already operating on multiple satellites, allowing CUS-GNC to collect telemetry and use it to refine its models and train neural networks for future missions. “The more it flies, the more it delivers,” Chesi says. “It is learning not just from simulation, but also with experience.”

From Simulation to Flight

Engineers can test individual spacecraft components in a laboratory, but reproducing the environment in which a complete GNC system must operate is much more difficult. Earth’s gravity prevents them from fully replicating the frictionless, six-degree-of-freedom motion that a spacecraft experiences in orbit.

That limitation makes simulation a critical part of the development process. “The moment they launch, that’s it,” Chesi says. “There is no way to go up there, pick up the satellite, bring it back, and do it again.”

CUS-GNC uses MATLAB and Simulink to design algorithms, model spacecraft dynamics, simulate missions, and evaluate SpacePilot’s response under different conditions. Aerospace Toolbox helps the team model the space environment and serves as a reference for validating its models.

After evaluating an algorithm in simulation, CUS-GNC generates code from its MATLAB and Simulink models and packages it as a library for the customer. The team runs the code on a representative onboard computer to evaluate timing, latency, input and output behavior, and resource use. The engineers then test how the software interacts with engineering models of the spacecraft’s sensors and actuators.

This workflow lets CUS-GNC compare results across simulation, representative hardware, customer testing, and spacecraft in orbit. Chesi says flight telemetry closely matches the team’s simulation results, giving engineers greater confidence in their models and helping customers evaluate SpacePilot for future missions.

The team also uses 3D visualization to replay telemetry and investigate spacecraft behavior. Visualizing data in a virtual environment can reveal how factors such as spacecraft orientation or the position of the sun affect behavior that may be difficult to interpret from telemetry values alone.

Scaling a Space Startup

For a small engineering team, Model-Based Design shortens the path from an idea to software that can operate onboard a spacecraft.

In a traditional workflow, one team may develop an algorithm in simulation before another team rewrites it as flight software. That handoff adds time and can introduce differences between the behavior engineers evaluate in simulation and the code they eventually deploy. By generating code from its models, CUS-GNC tests an implementation that more closely reflects the algorithms its engineers design and validate.

The workflow also lets the team spend more time on the engineering problems that distinguish SpacePilot. “All the time that we otherwise would have spent coding allows us to concentrate more on the algorithm and on the design,” Bianchi says.

The MathWorks Startup Program provides CUS-GNC with affordable access to the tools that its engineers use throughout the development process. The program also connects the startup with MathWorks engineers who help the team identify efficient approaches to new products, features, and technical challenges. According to Bianchi, that support helps CUS-GNC build confidence in its solution and demonstrate credibility with customers.

That credibility matters when a startup asks a customer to place mission-critical software on board a spacecraft. CUS-GNC can point to code operating in orbit, results that closely match its simulations, and customers using SpacePilot as part of their missions. What began as prototype software now operates as a production product.

CUS-GNC envisions SpacePilot becoming a reusable autonomy layer for spacecraft across a range of missions. One of the most meaningful signs of progress is already visible. “We are seeing that our work is translating to something that is actually useful for our customers,” Chesi says. “Our product is used today.”

Hear from Simone Chesi and Marco Bianchi of CUS-GNC:

Learn more about CUS-GNC: https://www.cus-gnc.com/ 

Learn more about MathWorks Startup Program: https://www.mathworks.com/products/startups.html