SAN FRANCISCO — In Orbit Aerospace won a $1.8 million AFWERX contract to employ artificial intelligence to detect and manage anomalies in hypersonic flight.
In Orbit is developing machine-learning tools to “enable more reliable flight operations over the entire range of flight envelopes,” CEO and co-founder Ryan Elliott told SpaceNews.
Primarily, In Orbit is focused on third-party logistics and infrastructure for in-space manufacturing and research. In Orbit plans to transport cargo to and from space stations.
Since return trips through Earth’s atmosphere will be conducted at hypersonic speeds, In Orbit engineers are eager to use artificial intelligence to make vehicles more reliable.
During Phase 1 of the AFWERX Small Business Technology Transfer (STTR) program, In orbit worked with the University of Colorado, Boulder, Center for National Security Initiatives (NSI) to create machine-learning algorithms to detect hypersonic vehicle anomalies as they were happening.
“One of the main challenges for hypersonic vehicles is that it’s nearly impossible for a human to make effective decisions in real time to mitigate any anomalies or undesirable conditions during flight,” said Ishaan Patel, In Orbit chief technology officer.
The Defense Advanced Research Projects Agency’s experimental hypersonic glide vehicle, Hypersonic Technology Vehicle 2, for example, reached an estimated speed of Mach 20 during a nine-minute test flight.
For the Phase 2 STTR, In Orbit and NSI will use machine learning “to not only detect anomalies and faults onboard the vehicle, but to manage those faults and make informed decisions to reduce the risk to the mission and the vehicle,” Patel said. If successful, the algorithms also will “reduce cost and increase mission success,” he added.
Finding Training Data
A challenging aspect of this initiative, like many AI-related projects, is finding data to train machine-learning models. Fortunately, NSI has expertise in identifying available datasets and creating training data, Patel said.
“We have an extensive catalog of data that we can tap into,” Patel said. “As part of this effort, we’re also going to be creating synthetic training data.”
Under another AFWERX contract, In Orbit and the University of Colorado, Boulder, are exploring spacecraft docking with electrostatic adhesion technology.
In Orbit, founded in 2020, conducts research and development in a 1,707-square-meter facility in Torrance, California.