Driving · 2026-11-03

An intelligent satellite must do nothing well

Fabrizio Pilone, Giordana Bucchioni

arXiv:2611.02922PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

The authors argue that algorithm choice is not the main design axis for autonomous space inspection; fuel efficiency depends on representation and planning choices that let an agent exploit natural drift (no rest state) and plan long enough. A control barrier function (CBF) filter eliminates keep-out-zone (KOZ) violations in evaluation episodes where learned policies alone can fail.

Why it matters

For orbital inspection, performance hinges on what the agent can represent (especially the consequences of drifting) and on planning horizons that reveal long-term benefits. The paper also shows that learned policies cannot be assumed safe by construction, and a safety filter is needed to guarantee KOZ constraint satisfaction during evaluation.

Method

  • Controlled ablation ladder varying one design axis at a time (information availability, state representation, algorithm choice, action space, planning horizon) under identical conditions for viewpoint-level and thrust-level agents.
  • Orbit inspection modeled with Clohessy-Wiltshire dynamics (no rest state): agents can either thrust or drift along natural trajectories for fuel-free motion/observations.
  • Safety ensured with a control barrier function (CBF) filter beneath the decision layer, posed as a QP using a KOZ outside set C defined by h(x) = sqrt(2 a_max(||r|| − R_KOZ) + v_p).

Limitation

The study states that learning policies cannot certify safety by typical episode behavior, and thus uses a CBF filter for evaluation; it does not provide a guarantee without the filter beyond reported episode statistics.

Abstract (from arXiv)

Autonomous inspection of a resident space object is usually posed as a contest between classical next-best-view planning and reinforcement learning. We argue that the choice of algorithm is the least informative of the design axes. An inspector spacecraft has no rest state: when it stops thrusting, orbital dynamics carry it along a natural trajectory that can be used to observe the target at no fuel cost. On a controlled ladder in which one design axis is varied at a time, fuel efficiency turns out to be set by three things: whether the agent is allowed to drift, whether its state records where drifting has taken it, and whether its planning horizon is long enough for the benefit to appear. Removing any one of them doubles the fuel bill or more. The algorithm still matters, but through what it can represent: a model-based agent whose learned world model ignores fuel spends several times more of it. Finally, no learned policy is safe by construction, and a control barrier function filter beneath the decision layer removed every keep-out-zone violation in two thousand evaluation episodes.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First announced by arXiv: 2026-11-03 · Indexed by Pub-AI: 2026-11-03 UTC

preprint

An intelligent satellite must do nothing well

Fabrizio Pilone, Giordana Bucchioni

First announced by arXiv: 2026-11-03 · Original record ↗

arxiv
2611.02922
Explicit source versions (1)
  • v1 · Date not supplied
  1. Preprint available

    Source date: 2026-11-03 · First detected: 2026-11-03 UTC · Historical baseline

    Source evidence (1)
    • Official evidence ↗ · RSS guid=oai:arXiv.org:2611.02922v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv

Autonomous inspection of a resident space object is usually posed as a contest between classical next-best-view planning and reinforcement learning. We argue that the choice of algorithm is the least informative of the design axes. An inspector spacecraft has no rest state: when it stops thrusting, orbital dynamics carry it along a natural trajectory that can be used to observe the target at no fuel cost. On a controlled ladder in which one design axis is varied at a time, fuel efficiency turns out to be set by three things: whether the agent is allowed to drift, whether its state records where drifting has taken it, and whether its planning horizon is long enough for the benefit to appear. Removing any one of them doubles the fuel bill or more. The algorithm still matters, but through what it can represent: a model-based agent whose learned world model ignores fuel spends several times more of it. Finally, no learned policy is safe by construction, and a control barrier function filter beneath the decision layer removed every keep-out-zone violation in two thousand evaluation episodes.