What Happened
On August 17, NASA announced that a small satellite in low Earth orbit had done something no spacecraft had done before: it worked out its own position by looking at other objects floating around it, with no GPS signal and no instructions from the ground. The experiment ran aboard Starling, a swarm of four CubeSats that NASA launched in 2023 to test what happens when spacecraft are allowed to coordinate among themselves. The technology demonstration carries a deliberately literal name — FALCON, for Fast Autonomous Lost-in-space Catalog-based Optical Navigation — and the "lost-in-space" part is the honest description of the problem it solves. Drop a satellite into orbit with no idea where it is, and FALCON is the software that lets it figure that out by itself.
The clever part is that FALCON did not need new hardware. Every spacecraft carries star trackers, small cameras that photograph the star field and match it against a stored map so the vehicle knows which way it is pointing. Starling's team pointed those same cameras at a different problem. The mission loaded a catalog of roughly 20,000 known space objects and their predicted orbits onto the spacecraft, then let FALCON match the moving points its cameras actually saw — other satellites, spent rocket bodies, fragments of debris — against that list. Once enough objects were identified and verified, they became surveyor's stakes. The satellite triangulated off its neighbors.
A second set of experiments ran the logic backwards. Instead of using the catalog to locate Starling, FALCON used Starling's observations to correct the catalog. Over a three-day stretch, it refined the orbits of more than 200 space objects without a single command from operators on the ground, and the position predictions it generated onboard came out sharper than the ones being supplied by ground stations. A CubeSat the size of a cereal box was quietly producing better orbital data than the network built to track it.
FALCON is a joint flight experiment between NASA and EraDrive, a startup spun out of Stanford University, combining EraDrive's Era-Core flight software and embedded algorithms with Starling's existing cameras. The partnership did not start as a commercial deal. It began as a University SmallSat Technology Partnerships project — a research grant — and grew into a company now selling the software it proved in orbit. NASA's Ames Research Center in Silicon Valley leads the Starling mission, funded through the agency's Small Spacecraft and Distributed Systems program.
Why It Matters
GPS has become such reliable plumbing that it is easy to forget it is a service, not a law of physics. Satellites in low Earth orbit lean on it constantly, and it works because a constellation of dedicated navigation satellites sits above them broadcasting timing signals downward. Climb higher and the geometry falls apart. Head for the Moon and the signal is effectively gone. Every mission that wants to operate past that boundary has historically solved the problem the expensive way: by asking Earth. Deep Space Network time is scarce, round-trip light delay is real, and a spacecraft that cannot navigate without a phone call home is a spacecraft on a leash.
That leash is the constraint NASA is trying to cut. The agency's stated plans involve satellite swarms around the Moon, distributed science missions where a dozen instruments take simultaneous readings from different points in space, and eventually surface operations on the Moon and Mars that need positioning support overhead. All of those depend on spacecraft knowing precisely where they are relative to one another. Building a lunar GPS constellation from scratch is one answer. Teaching spacecraft to navigate by whatever happens to be nearby is a cheaper one, and it degrades gracefully — there is no single point of failure to knock out.
The near-Earth payoff is arguably more immediate. Orbital congestion has outrun the ground-based tracking systems built to manage it, and the accuracy of the public catalog is now a genuine safety variable: collision-avoidance decisions get made on predicted positions that may be hours stale. A network of satellites that continuously observe their neighbors and improve the catalog in flight turns every spacecraft into a sensor rather than just an object to be tracked. Roger Hunter, program manager for NASA's Small Spacecraft and Distributed Systems program at Ames, framed the implications around exactly that: space-traffic monitoring, collision avoidance, and alternative navigation.
It is also worth being precise about what kind of autonomy this is. FALCON is not a large model reasoning about space. It is catalog matching, pattern recognition, and orbit estimation compressed into embedded software small enough to run on a CubeSat's modest processor and reliable enough to trust without supervision. That is a useful counterweight to the prevailing story about machine intelligence, which tends to measure progress in parameter counts and datacenter megawatts. The work that actually leaves the lab is often the work that fits in a shoebox and never needs a network connection.
Reaction
NASA's own framing was matter-of-fact. Hunter called FALCON "yet another success for the Starling demonstration mission," adding that the number of firsts coming out of the swarm keeps growing — a reasonable claim for a mission that has already demonstrated autonomous swarm maneuvering and inter-satellite networking on the same hardware.
The space press seized on the phrase that NASA had buried in the acronym. Coverage across outlets from Space.com to SatNews led with the "lost-in-space" test, and the result surfaced again more than a week later as a lead item on ScienceDaily, which is unusual staying power for a CubeSat software demo. Within the space situational awareness community the catalog-improvement result drew the closer attention. Commercial constellations have multiplied faster than tracking capacity, and a method that recruits those constellations into doing the tracking is a structural answer rather than an incremental one.
What's Next
Starling is not finished. Later this year the mission will extend the FALCON experiment using Era-Core across the full swarm, letting all four spacecraft share tracking observations and refine their positions collectively rather than individually. That is a meaningfully harder problem — the satellites have to agree on a shared picture of where everything is, including themselves, while each holds only a partial view — and it is the version that actually maps onto a lunar swarm.
On the commercial side, EraDrive is packaging Era-Core and related hardware for wider use, which is the outcome NASA's small-spacecraft program is designed to produce: a university research thread that turns into flight-proven software someone else can buy. For NASA, the target applications are the ones already on the schedule — coordinated satellites supporting crewed operations on the lunar surface, distributed science platforms that need every instrument's position pinned down before the measurements mean anything, and traffic management in an orbital environment that keeps getting more crowded.
Closing Thoughts
What stays with me about FALCON is how little of it is new. The cameras were already there. The catalog was already public. The orbital mechanics have been understood for centuries. What changed is that someone wrote software good enough to combine those three things onboard, in real time, without asking permission — and then flew it to see whether it held up. The result is a spacecraft that is less dependent on the ground than the one launched three years ago, running on identical hardware.
There is a version of autonomy that gets talked about constantly and a version that gets shipped. The first is about capability: what can the system do that it could not do before. The second is about dependence: what can the system stop needing. FALCON is firmly the second kind, and it is the kind that tends to matter when the ground station is 384,000 kilometers away and the answer to every question takes a second and a half to arrive. A satellite that can locate itself among strangers is a small thing. It is also, quietly, the precondition for most of what comes next.
한글 요약
NASA는 8월 17일, 소형 위성 군집 미션 '스탈링(Starling)'이 GPS 없이 스스로 궤도상 위치를 계산하는 데 성공했다고 발표했습니다. FALCON(Fast Autonomous Lost-in-space Catalog-based Optical Navigation)이라 불리는 이 실험은 위성에 이미 달려 있던 별 추적 카메라를 그대로 활용합니다. 약 2만 개의 알려진 우주 물체 목록을 위성에 미리 실어 두고, 카메라에 잡힌 다른 위성과 우주 쓰레기를 이 목록과 대조해 기준점으로 삼은 뒤 자신의 궤도를 역산하는 방식입니다.
반대 방향 실험도 함께 이뤄졌습니다. 스탈링의 관측 데이터로 목록 자체를 보정한 것인데, 사흘 동안 지상 운영진의 개입 없이 200개가 넘는 물체의 궤도 정보를 개선했고, 위성이 자체적으로 산출한 위치 예측이 지상국이 제공한 값보다 더 정확했습니다. 이 소프트웨어는 스탠퍼드대에서 분사한 스타트업 에라드라이브(EraDrive)와 NASA 에임스 연구센터가 함께 만든 것으로, 대학 소형위성 연구 과제로 시작해 상용 제품으로 이어진 사례입니다.
의미는 두 방향입니다. 달이나 심우주에서는 GPS 신호를 쓸 수 없기 때문에, 자율 항법은 달 궤도 위성군과 유인 탐사를 위한 전제 조건에 가깝습니다. 동시에 지구 저궤도에서는 위성 하나하나가 관측 센서가 되어 우주 물체 목록을 실시간으로 갱신할 수 있어, 충돌 회피와 우주 교통 관리에도 곧바로 쓸 수 있습니다. 거대 모델이 아니라 큐브샛에 들어갈 만큼 작은 임베디드 알고리즘이 만들어낸 결과라는 점도 눈여겨볼 만합니다. NASA는 올해 안에 네 대의 위성이 관측 데이터를 서로 공유해 위치를 함께 보정하는 확장 실험을 진행할 예정입니다.