Our work "Learning in Orbit: Onboard Training and Closed-Loop Attitude Control with a Thermal Segmentation Model on a Spacecraft" is finally on arXiv
arxiv.org/pdf/2610.19122
In 2025, we demonstrated both learning (training) and inference of a vision model onboard a spacecraft, running on an
@nvidia Jetson TX2 NX, and used it for closed-loop attitude control. For me, the motivation was to show that the algorithms & compute we develop in robotics on Earth can fly in space.
I may be biased, but I think this paper is especially valuable to the community. Beyond the algorithm innovations, we document the open-source stack that made it work in orbit, including ROS 2 (
@OpenRoboticsOrg) and
@OpenZFS, along with the failure cases and lessons learned along the way.
Joint work with Hannah Grauer (
@hannah_grauer), Phaedrus Leeds, Nikhil Ranganathan, Connor Lee, Sulekha Kishore, Cailyn Smith, Shon Cortes, Brian Hardy, Benjamen Bycroft, Darren W. Rowen, and Soon-Jo Chung.
@AerospaceCorp
I attached my favourite thermal picture: moon rise.