IBM open-sources AI for optimizing satellite communications and predicting space debris trajectories

IBM today announced two new open source initiatives aimed at solving the technological challenges surrounding cube satellites, a form of space science miniaturized satellite, and understanding of space situations. Both were constructed by the Space Technology Hub team of the organization and are available on IBM’s Red Hat OpenShift platform as containerized deployments. 

The Space Tech Hub team, led by Naeem Altaf, IBM’s distinguished engineer and space technology CTO, also partners with state agencies, colleges, and space technology companies to pioneer spacecraft and satellite solutions. The team today unveiled KubeSat, an autonomous structure designed for tiny satellites that supports connectivity optimization. They also unveiled the framework of Space Situational Awareness (SSA), which is based on a series of algorithms that decide where human-constructed objects are positioned orbiting the Earth and might drift in the future.

“Our team is committed to fostering the future of creativity in space. IBM wrote in a blog post that with the proliferation of satellites of all sizes in lower Earth orbit in space, huge amounts of data would be generated related to Earth observation, space traffic control, space situational awareness.

KubeSat

As a “cognitive” suite designed to simulate orbital mechanics for objects through Orekit, a low-level space dynamics library written in Java, IBM describes KubeSat. KubeSat uses calculations to limit communications between satellites, ground sensors, and more by integrating NATS messaging systems, optimizing machine learning communications, and publishing them on a dashboard. (NATS is a high-performance messaging system that functions as a queue of distributed messaging.) IBM says it is possible to use KubeSat to model how clusters are shaped by cube satellites and communicate with items such as ground stations. 

The architecture of KubeSat allows users to swap unique use cases into AI models. Communications can be autonomous between satellite swarms, allowing swarms to integrate or separate as necessary.

IBM is open-sourcing the app code for Orekit alongside KubeSat, which it has developed with undergraduate students at Stanford University. Extensions from Orekit help simulate communications from satellite to satellite, satellite to ground station, and satellite to ground sensor over a meshed NATS messaging network, optionally with a custom visualization layer.

Space Situational Awareness (SSA)

The open-source SSA framework was created today jointly with Professor Dr. Moriba Jah of the University of Texas, Austin. SSA prevents the pitfall of trying to craft a model to predict all the orbital dynamics of an object by using AI techniques to determine when physical models wrongly predict the potential position of an object. 

Tens of thousands of human-constructed structures are known to circle the Earth, flying at speeds reaching 8,000 meters per second. Problematically, position data about these objects appear to be uncommon and noisy, and in unpredictable ways, factors such as space weather and atmospheric density influence their trajectories.

To train baseline AI models, SSA sources data from the United States Strategic Command. The dataset is updated about once a day for most objects in low-Earth orbit, IBM says. A physics-based model forecasts disruptions caused by the Earth in the orbit of an object, while a machine learning model anticipates mistakes in the predictions of the physical model. SSA blends the two models and changes the prediction of the physical model according to the expected error. 

“A defining concern of the ‘space 2.0 era’ is the simulation and impact prediction of anthropogenic space objects,” IBM added. “You can now enter our project to solve the ‘space garbage’ problem by open-sourcing a complete pipeline to make orbital forecasts and display the results.”