SpaceX may be approaching debris detection as a machine-learning problem in which the entire constellation, not individual satellites, is learning to avoid collisions.

SpaceX delayed last Wednesdays Starlink launch due to high winds and on Thursday they decided to do a software update and postpone the launch until next week, but they revealed significant progress in their Starlink mission press release and in tweets by and a media call with Elon Musk.

Starlink size comparison – novel packaging accommodates 60 satellites in a single launch. (Source)

The mission press release said SpaceX has significantly reduced the size and weight of their satellites. Their initial November 2016 FCC filing specified 386 kg satellites that measured 4 x 1.8 x 1.2 meters. In February 2018, they launched two Internet-service test satellites—TinTin A and B—that measured only 1.1 x .7 x .7 meters with a total mass of approximately 400 kg. The mass of the Starlink satellites will be only 227 kg, about 43% that of the test satellites. (They are still heavier than OneWeb’s 147.4 kg test satellites)

As far as I know, SpaceX has not previously commented on the number of satellites that might be launched at once, but the number was generally estimated as 25-30 after considering constraints on mass, volume, and numbers of satellites per orbital plane. As shown here, they will be launching a surprising 60 flat-packed satellites. Launching 60 satellites also demonstrates continued progress in rocket capability—this will be the heaviest SpaceX payload ever.

The speed and density of satellites in
low-earth orbit increase the likelihood
of a cascading debris collision. (Source)The current and planned proliferation of low-earth orbit satellites increases the likelihood of a Kessler Syndrome event—a cascade of collisions between satellites and the ensuing debris. The press release alluded to what may be a significant advance in debris mitigation, stating that:

Each spacecraft is equipped with a Startracker navigation system that allows SpaceX to point the satellites with precision. Importantly, Starlink satellites are capable of tracking on-orbit debris and autonomously avoiding a collision.

That would be a breakthrough if feasible, but on first consideration, it seems impossible. Low-earth orbit satellites move very fast and even if a satellite had the resolution and pattern-recognition capability to “see” debris in its path, it would not be able to maneuver quickly enough to avoid a collision. That point was raised in this online discussion and a possible solution suggested—the entire constellation could dynamically pool and share data from each satellite as well as use NORAD tracking data, which Musk mentioned during the media call.

SpaceX may be approaching this as a machine-learning problem in which the entire constellation, not individual satellites, is learning to avoid collisions using its shared data as well as data from other sources like NORAD. One can imagine sharing such data with competitors like OneWeb and Telesat or even with Russia, China or India. (Elon Musk is known to read science fiction—this speculation is reminiscent of Azimov’s Gaia or Teilhard de Chardin’s noosphere).

The prospect of launching 60 satellites at once and a shared-data approach to collision avoidance have grabbed my attention, but Musk’s tweets and media call were also highly informative—a few examples:

• There’s a lot of new technology in the satellites—things could go wrong.
• The initial constellation will not have inter-satellite links.
• Each Starlink costs more to launch than it does to make.
• They will partner with telephone companies in sparsely populated regions.
• Six more launches of 60 sats needed for minor coverage, 12 for moderate.
• We have sufficient capital to build an operational constellation.

All that and they have yet to launch the satellites—stay tuned.