← Founder Notes
Note 01 · space / reinforcement learning / autonomy

The orbit is now being asked to police itself

Yethikrishna ROriginal on Threads

the orbit is now being asked to police itself: a reinforcement learning policy trained with ppo dodged debris in 97.5 percent of geostationary scenarios where a rule based controller managed 20.7, uk’s project nova watches the growing wreckage, and satellites now plan maneuvers on board without waiting for ground commands.

the orbit now files its own traffic report.

Context

An arXiv paper proposes a reinforcement learning policy for autonomous collision avoidance trained with Proximal Policy Optimization in an open-source astrodynamics simulator, and reports that in 1,000 deterministic GEO episodes the agent achieves a 97.5 percent collision avoidance success rate against 20.7 percent for a rule-based baseline and 27.5 percent for an impulsive delta-v planner. The simulator uses real-world and simulated debris.

The UK Space Agency says Project NOVA plans 13 optical telescopes across five countries to monitor satellites, orbital debris and near-Earth objects independently. It says about 14,000 satellites are active and as many as 100,000 are expected by 2030. A separate arXiv paper by Bourriez, Loizeau and Abdin studies reinforcement learning for spacecraft autonomous decision-planning for collision avoidance.

How it compares

The 97.5 and 20.7 percent figures match the arXiv abstract, with two qualifiers. They come from 1,000 simulated, deterministic GEO episodes, not from real satellites, and the paper reports a second baseline at 27.5 percent.

Project NOVA is a ground-based telescope network that tracks hazards and gives early warnings, according to the UK Space Agency, and it is a separate effort from the learned policy. The note joins them as one story, and the sources do not connect them. The word 'watches' fits NOVA only as a plan, since the first site was visited in May 2026 and the 13 telescopes are planned.

Satellites planning maneuvers on board without waiting for ground commands is what the autonomous decision-planning paper studies as a research approach. The pages read do not show an operational satellite doing this, so a deployed system is unsupported here, not refuted.

'The orbit now files its own traffic report' is the author's line.

Related work

Watch next

  • Check whether any operator has flown onboard autonomous collision avoidance. Read the simulator's validation against real conjunction data.

Sources

  1. arXiv: Satellite Trajectory Optimization via Proximal Policy Optimization for Space Debris Avoidancearxiv.org

Provenance

The note above is reproduced unedited from the original post, first published on Threads on 3 October 2026 at 18:06 IST. Sources are the papers and datasets the note draws on.

View the original post
Embed this note
<iframe src="https://founder.myndlabs.tech/notes/embed/the-orbit-is-being-asked-to-police-itself" width="480" height="420" style="border:0;max-width:100%" loading="lazy" title="The orbit is now being asked to police itself"></iframe>

More notes