Ottawa’s Obruta Space Solutions will be putting their autonomous spaceflight technology to the test next month on the International Space Station (ISS).
Their systems for spacecraft vision and for guidance/navigation, funded in part by a pair of new Space Technology Development Program (STDP) awards, will be tested using the International Space Station’s “Astrobee” drones.
Obruta Co-founder and CEO, Kevin Stadnyk, gave more details in an interview with SpaceQ.
Obruta is a spacecraft guidance company that pivoted from debris removal, after successful participation in the University of Toronto’s Creative Destruction Lab (CDL) Space Stream. Their time at the CDL led them to realize that they needed to target a more directly monetizable market segment than space debris. While they were interested in the debris question, the question of “who will pay for debris removal” proved potentially insurmountable.
So, instead, they shifted their focus to providing Rendezvous, Proximity Operations, and Docking (RPOD) capabilities for space companies. Calling their RPOD kit “the best spacecraft pilot in the universe,” they are focused on providing spacecraft the capability to autonomously dock with other spacecraft, allowing for spacecraft companies to perform in-orbit servicing, refuelling, orbit correction, and logistics.
Stadnyk said that they believe “these are critical aspects of the in-space economy,” and that developing them “will support rapid growth.” In particular, Stadnyk highlighted the possibilities involved in space-based manufacturing, and how autonomous spacecraft will be key for delivering materials and finished products to space-based manufacturing infrastructure.
STDP Awards
The Canadian Space Agency (CSA) appears to agree with the potential of autonomous spacecraft. They provided two STDP awards to Obruta this year. The first, for $222,398, is to help Obruta develop their “Computer Vision for Autonomous Spacecraft Docking” according to the STDP announcement. The second, for $222,381, is aimed at developing their “Guidance, Navigation, and Control for Autonomous Spacecraft Docking” capabilities according to the same announcement. Both are part of the STDP’s “Space Research and Development: Small Businesses” category.
Stadnyk elaborated on the awards. He said that they both interrelated parts of the larger RPOD project. The first award, for computer vision capabilities, is “the first step of the RPOD software and hardware kit.” It is the part of RPOD that ingests the information from the spacecraft’s sensors, fuses it, determines the state of the objects around it, and then passes that information along to the guidance and navigation systems.
The second award is for the development of the guidance and navigation systems. Those systems take that information and “calculate the trajectories that must be executed to get from point A to point B, aligned with the spacecraft goal of ultimately docking with another spacecraft in orbit.” Those instructions are then passed along to the control system, which determines the thruster firings and maneuvers needed to achieve those trajectories.
Stadnyk said that machine learning/artificial intelligence (ML/AI) is a key part of the first award’s research and development. ML/AI is used heavily for the computer vision stage of the process, as it’s a well-known and generally trusted method for distilling object information from optical data, and he has people on his team (including his co-founder, Kirk Hovell) that have strong machine learning backgrounds.
When it comes to the guidance and navigation part, however, they’re taking a more traditional approach, with much less use of ML/AI.
Stadnyk noted that the space environment isn’t as crowded and chaotic as the terrestrial conditions that are making automotive autopilot so challenging to implement. He also doesn’t believe that the “hallucinations” plaguing large language model (LLM) AIs like ChatGPT would affect attempts to deploy ML/AI-based tools in orbit.
Stadnyk said that they’re simply “sticking with known software solutions where that is the best option,” adding that “there’s no reason to change what’s not broken.” As “space dynamics are quite well understood,” he said, it makes more sense to deploy tried-and-tested code for the guidance, navigation, and control tasks. This is especially important as ML algorithms can be a “black box,” where errors are difficult to diagnose and correct.
So, at the moment, they’re saving ML/AI deployment for the computer vision task where it actually outperforms traditional solutions.
Astrobees
Their other recent announcement was that they’ll be able to put their RPOD tech to the test starting next month.
In their announcement on Linkedin, Obruta said that they’ll be using “Astrobees” for testing. The Astrobees are cube-shaped robots on the ISS that are used to help astronauts with routine tasks like taking inventory and moving cargo. Obruta will be borrowing them from the ISS astronauts to perform tests.
Stadnyk elaborated on the announcement. He said that they were awarded this time by the ISS National Lab’s program, and will be working in cooperation with an American company GeoJump. Stadnyk demurred when asked about GeoJump’s specific role, though.
He said that they will be testing “the full suite of the guidance, navigation and control software,” as well as doing a number of computer vision tests. Obruta will be looking at “how our algorithms and software are able to detect and track the other Astrobee robots floating around the one with our software.” It will also be attempting various maneuvers, including “testing different types of trajectories” like flying circles around other Astrobees, performing a simulated inspection maneuver, or checking appropriate and safe docking speeds.
Obruta will be using the Astrobees, Stadnyk said, “the same way we would test any other system in space during a full scale mission.”
Stadnyk said that the testing will be spread out across the first three months of 2024. The first day of testing will be happening “no earlier than January 15th.” Obruta will take a month to download the data, analyze the findings, and prepare for the next test. They’ll return to the Astrobees a month later in February for more testing, and will repeat the whole process before a final testing day in March.
Once the Astrobee testing is done, Stadnyk said, they’ll have a “full view of the data and test results.” With that data, they’ll be refining their systems, and “make more improvements as we push towards a commercially ready solution in the back half of 2024.”