Satlyt completes orbital planning test on one computer, not a satellite network

Satlyt wants satellites to share computing work, but its latest milestone stops short of that goal. The startup said software running on one computer in orbit generated plans for distributing tasks across a simulated four-computer network. The October test returned its outputs to Earth; it did not demonstrate separate spacecraft exchanging data or sharing computing jobs.

Completed after 13 months of work, the experiment assessed how to assign computing tasks based on available processing capacity and communications. Founder and CEO Rama Afullo described the results to TechCrunch, which published the account on October 30. It was a test of one planning component, rather than the full system Satlyt hopes to build.

Four computers in the scenario, one in orbit

The simulated network contained four computers, four users and eight services. Only one physical computer aboard a spacecraft ran the software; the other three computers were simulated. Within that setup, the software evaluated where tasks should run while accounting for both computing resources and communication availability.

Satlyt’s intended system would let users set permissions, priorities and operating limits, then leave the software to make decisions inside those boundaries. That is the proposed division of responsibility: people define what is allowed and important, while the software decides how to allocate the work.

The larger aim is to run AI and other applications closer to where satellites collect data, reducing reliance on sending readings to Earth for analysis. Afullo offered wildfire detection as an example: onboard processing could help a spacecraft send useful information sooner, rather than depending entirely on ground-based interpretation.

This experiment tested one planning component of that broader capability.

Rama Afullo, Satlyt founder and CEO, speaking to TechCrunch

A planning tradeoff and a timing error

Afullo said the experiment revealed a tradeoff between planning approaches. One worked best but required more computing power than the alternatives. The result therefore was not simply that the software could produce a plan: its choice of approach also carried a resource cost on the hardware doing the planning.

The team also found a problem in how the software represented communication delays. It recorded arrival time rather than elapsed delay—the difference between when information arrived and how long it took to get there. Afullo acknowledged the issue while saying that software development, onboard execution and the return of outputs had succeeded.

The next demonstration needs separate spacecraft

The research was conducted with the University of Houston using NASA funding. That backing is not a deployment commitment: NASA has not selected Satlyt for a particular mission or chosen to deploy the technology. Satlyt also says its software is already aboard three satellites, though this experiment used only one physical computer.

Moving from generated plans to shared computing will require more than adding machines. Afullo identified several requirements for operating the proposed network:

• Usable communication links and permission from participating satellite operators.
• Secure data transfers and scheduling of available computing resources.
• Recovery when a connected group of computers becomes unavailable.

Satlyt’s next steps are broader validation and integration, including tests under different workloads, hardware and connectivity conditions. Demonstrations of computing work shared across separate spacecraft would follow. Afullo’s longer-term goal is to let developers build space applications without rebuilding the underlying infrastructure for every mission; those cross-spacecraft demonstrations remain ahead.