Future Mission Enablers: The Advanced Technologies Optimizing Space Industry Manufacturing and Systems Operations
During last month’s popular webinar Exploring the Benefits of Artificial Intelligence (AI) for Future Spacecraft, speakers from Satcoms Innovation Group (SIG), NASA Ames Research Center, and Ball Aerospace highlighted the use of artificial intelligence to optimize communications, spectrum management, LEO constellation operations, and crew assistance. We received so many questions from viewers that we couldn’t answer all of them during the actual webinar, but speakers Dan Reagan (mission systems analyst at Ball Aerospace) and Roberto Carlino (software/hardware test engineer for Astrobee Free-Flying Robots at NASA Ames Research Center) reviewed your questions after the event and have given you answers on artificial intelligence and Astrobee below.
Isn’t AI just a fancy auto-generation of if/then statements?
Dan: To the degree that a computer is just a fancy auto-generation of ones and zeros, or that language is just an arrangement of sounds or scribbles, yes – fundamentally.
What is the proof that AI will help to optimize space mission resources?
Dan: I published a paper last year which demonstrates an initial, quantitative proof-of-concept of a modular neural reinforcement network's ability to optimally task a space situational awareness system. The integrated network demonstrated performance improvement against mission-relevant metrics even after a relatively small number of training epochs. You can find the paper here.
How do you envision AI be used in flight software with real-time response requirements?
Dan: While training an AI system can be a long, arduous process, the deployed code is parallelizable, accelerable, and quick in general. I foresee – at least at the beginning – little need to train neural networks on-orbit. On the ground, where processing is nearly unlimited, AI can live in the cloud and continually adapt and learn in real time. In space, I would expect flight code to be static rather than living, in which case we can expect performance not unlike the flight code we execute in space today.
With regards to Astrobee or any other area of this material: What sort of real-time operating system (RTOS) features for AI are being uniquely developed, tested, and certified? Any unique approaches for AI within RTOS for safety and/or flight certs within the given environment?
Roberto: We do not use any RTOS features, and thanks to our architecture it is not required. More info is available in a paper by Lorenzo Fluckiger, Kathryn Browne, Brian Coltin, Jesse Fusco, Theodore Morse, and Andrew Symington: Astrobee Robot Software: Enabling Mobile Autonomy on the ISS”, In Proc. of the Int. Symposium on Artificial Intelligence, Robotics and Automation in Space (i-SAIRAS), 2018.
What are tasking problems that AI could help with?
Dan: At least initially, the best role for AI in tasking is to decide between various “good” options. By “good”, I mean safe, valid, and beneficial, possibly even auxiliary. Building trust in AI systems is going to take some time, so the first step is allowing AI to decide which data to gather between known sources, and to gather data which might not have otherwise been gathered.
Good tasking applications would be SSA, commercial earth imaging on small EO satellites, and – like Astrobee – health monitoring systems onboard spacecraft. These three examples provide supplementary data to a larger database, cause little waste if the data is of no use, and potentially massive gains if the data proves to be fruitful.
How will maintenance be completed on the NASA Astrobees if it is required? Are all the astronauts on the ISS trained to provide maintenance and repair?
Roberto: The Astrobees are designed in such a way that if some parts or components need to be repaired, the astronauts will be able to make the repair or replacement under our supervision and guidance. The astronauts will be trained a few days before the required repair, and the replacement parts will be already stored in orbit on the ISS.
Can we already process large amounts of data onboard with the very limited process power we have today on satellites?
Dan: No. The fact is the current AI renaissance owes its existence to GPU computing, and the radiation testing of GPUs for orbital use is only in its infancy stage. GPU computing in space is entirely an experimental endeavor at this point, although I expect that to change within the next five to 10 years. Until we better understand how to use these processors in space, our ability to process large matrices of data quickly in space will be hindered.
With regards to Astrobee, the Simultaneous Localization and Mapping (SLAM) features are fairly rudimentary (e.g. point clouds only). Do you plan on leveraging some 10-15 years of SLAM approaches on terrestrial/flight systems? Also: Do you have more information available on the supervisory control approach?
Roberto: Astrobee does not use any SLAM features. It localizes with monocular vision and an inertial measurement unit (IMU) without any environmental modifications. Visual features detected on a pre-built map, optical flow information, and IMU readings are all integrated into an extended Kalman filter (EKF) to estimate the robot pose. More info here is available in the paper by Brian Coltin, Jesse Fusco, Zack Moratto, Oleg Alexandrov, and Robert Nakamura: Localization from Visual Landmarks on a Free-Flying Robot”, In Proc. of the Int. Conf. on Intelligent Robots and Systems (IROS), 2016.
Can Astrobee dock and charge itself? If the ISS had to be evacuated for a short time could the Astrobee continue to operate? It would be helpful to continue to ‘operate’ the ISS until the environment improved for the astronauts.
Roberto: Yes! Astrobee can dock and charge itself autonomously. If there was an evacuation of the ISS, theoretically the Astrobees could still operate since they are completely autonomous and don’t need any supervision of the astronauts.
Will the Astrobee system assemble anything in space, outside the ISS?
Roberto: Astrobee is only designed to work inside the ISS. Its propulsion and navigation systems allow it to be used only for IVAs.
Image credit: NASA Ames Research Center
Space Tech Expo 2019 (May 20-22, Pasadena, CA) will highlight exciting new and advanced technologies that are currently being implemented in the space industry, such as advanced robotics, autonomous systems, additive manufacturing, blockchain technology, and many more.
Keen to learn more on autonomous systems, artificial intelligence, and advanced robotics for space? Join Roberto Carlino at Space Tech Conference in Pasadena, CA as he speaks in the session Optimizing Mission Agility with Autonomous Systems: The Benefits and Challenges of Implementing Artificial Intelligence on Wednesday, May 22.