Open Source AI Models Tackle Space Junk Problem

According to a recent announcement from IBM, as reported by TechHQ, open-source AI is being employed to solve problems in outer space, tackling issues relating to space junk and satellite communication.

IBM has been creating computer technology to drive space exploration and communication ever since the 1940s, but now IBM will be using artificial intelligence to handle those tasks. IBM is working on two different space-related projects: KubeSat and SSA (Space Situational Awareness). KubeSat is intended to enable the creation and control of tasks for satellite swarms, while SSA is intended to track the position of space junk in Low Earth Orbit.

The two projects were recently unveiled by the Space Tech Hub Team at IBM. The leader of the space tech team is Naeem Altaf, and according to Altaf the KubeSat project is an autonomous framework that provides the tools needed to create and manage tasks for satellite swarms and constellations. Beyond this, the KubeSat framework can simulate communications between satellites, helping engineers optimize these communications.

As more and more satellites are launched, communication between the satellites becomes increasingly complex, needing to be automated and optimized. The framework employs machine learning algorithms to optimize communications between satellites, placing restrictions on communications between certain satellites. KubeSat could be used to simulate how cube satellites interact with ground stations even as automated communications occur between swarms. The communications are published on a web dashboard for others to see. KubeSat runs its simulations through Orekit, which is a dynamic library created in Java.

The KubeSat project was made open source in the hope that the satellite swarm industry could be democratized, enabling startups and swarm operators to make use of the emerging technology.

The SSA project is the result of a collaboration between the IBM Space Tech Hub team and Dr. Moriba Jah of the University of Texas. The goal is that the AI models can improve orbit predictions for objects in Low Earth Orbit. Low Earth Orbit is full of space debris, much of which are artifacts leftover from rocket launches, or decayed satellites. These objects orbit the Earth moving at thousands of meters per second and their trajectories can be suddenly altered by fluctuations in atmospheric weather and density. The orbit of these objects needs to be predicted, so that collisions between space debris and important space tech devices don’t occur. It's hoped that AI models can improve the orbit predictions.

The SSA models were trained based on data collated by the United States Strategic Command. The dataset is updated once every day. A physics model is used to generate initial predictions about the orbit of most objects in Low Earth Orbit, and a machine learning model is then used to predict errors in the physics models. The SSA combines the two models together to update the physical orbit model. The second model is a gradient boosting model based on XGBoost.

Much like KubeSat, the SSA models were made open-source in an attempt to encourage data sharing and collaboration between different space companies and tech companies. After all, satellite communication problems and space junk are a threat to everyone operating in space.

Both KubeSat and OpenShift have been made available through IBM’s Red Hat OpenShift platform.