On-board processing that turns frames into answers.
A small satellite can collect far more pixels than it can send down. Our edge compute module does the work in orbit: it registers, stacks and geolocates the imagery, and attaches intelligent metadata before anything is downlinked.
The downlink is the bottleneck.
A satellite sees a ground station for a few minutes per pass, and every raw frame has to share that window. Processing on board changes what has to go down: a finished, geolocated tile and its metadata in place of a stack of raw sub-frames.
Compact FPGA, ARM and GPU modules have made this practical on small satellites, and several operators now run AI inference in orbit.
Illustration: the compute module in place inside a 6U satellite.
Three jobs, in order.
- Register
- Aligns each short sub-frame using the star camera's attitude as the prior, so motion during a slew is known.
- Stack
- Adds the aligned sub-frames into one sharp image. The signal adds up faster than the noise does.
- Geolocate and describe
- Places the tile on the ground on board and attaches intelligent metadata to it.
Flown to 42 km on Stratos-1.
The Jetson compute module flew to 42 km with our cameras on Stratos-1, our first stratospheric flight, in May 2026. Intelligent imaging is validated on Stratos-2 in October 2026, and the first on-board stacking flight is planned for 2027.
Satellite edge computing FAQ
What is edge computing on a satellite?
Running the processing on the spacecraft itself, next to the sensor, so that results come down in place of raw data.
Which processors do you use?
Stratos-1 flew an NVIDIA Jetson compute module to 42 km in May 2026. Our orbital design gives the star camera and the main imager each a dedicated FPGA + ARM processor, with a separate flight computer running the spacecraft.
Does it work with any camera?
It is designed as part of our payload, where the star camera's attitude is the prior for stacking. Talk to us about other sensors.
Fly a pathfinder.
Reserve a star-camera engineering unit for 2027.