TY - GEN
T1 - Physically-Realizable Adversarial Attack against Spacecraft Pose Estimation via Active Optical Jamming
AU - Yin, Yifei
AU - Ji, Fengbiao
AU - Li, Dongyu
AU - Xu, Jun
AU - Guo, Pengyu
N1 - Publisher Copyright:
© 2026 IEEE.
PY - 2026
Y1 - 2026
N2 - Vision-based spacecraft pose estimation is the cornerstone of autonomous on-orbit servicing and non-cooperative rendezvous. Currently, Hybrid Modular Pose Estimation (HMPE) pipelines remain the industry standard for these safety-critical missions due to their interpretability. However, these systems fundamentally rely on the axiom that high-contrast image features correspond to static, rigid geometric structures. We illustrate that this assumption can be physically violated on spacecraft platforms via Physically-Realizable Active Geometric Jamming. We propose an adversarial strategy where the target spacecraft employs onboard steerable spotlights to project beams onto its own planar surfaces, such as solar arrays. These projections create artificial high-contrast features that lack consistent 3D geometric causality, acting as high-confidence outliers that overwhelm the rejection capabilities of RANSAC. To identify effective illumination configurations within the non-differentiable and periodic parameter space, we formulate the attack as a zeroth-order optimization problem. We propose a Topology-Aware CMA-ES that explicitly handles the cyclic nature of angular search spaces, initialized via a Spatially-Structured Adversarial Atlas. Simulation results, conducted in a high-fidelity space environment incorporating HDR rendering and optical bloom artifacts, confirm that this approach systematically degrades the inlier ratio of standard HMPE pipelines, exposing a critical vulnerability in vision-based spacecraft navigation.
AB - Vision-based spacecraft pose estimation is the cornerstone of autonomous on-orbit servicing and non-cooperative rendezvous. Currently, Hybrid Modular Pose Estimation (HMPE) pipelines remain the industry standard for these safety-critical missions due to their interpretability. However, these systems fundamentally rely on the axiom that high-contrast image features correspond to static, rigid geometric structures. We illustrate that this assumption can be physically violated on spacecraft platforms via Physically-Realizable Active Geometric Jamming. We propose an adversarial strategy where the target spacecraft employs onboard steerable spotlights to project beams onto its own planar surfaces, such as solar arrays. These projections create artificial high-contrast features that lack consistent 3D geometric causality, acting as high-confidence outliers that overwhelm the rejection capabilities of RANSAC. To identify effective illumination configurations within the non-differentiable and periodic parameter space, we formulate the attack as a zeroth-order optimization problem. We propose a Topology-Aware CMA-ES that explicitly handles the cyclic nature of angular search spaces, initialized via a Spatially-Structured Adversarial Atlas. Simulation results, conducted in a high-fidelity space environment incorporating HDR rendering and optical bloom artifacts, confirm that this approach systematically degrades the inlier ratio of standard HMPE pipelines, exposing a critical vulnerability in vision-based spacecraft navigation.
KW - CMA-ES
KW - Data-Driven Optimization
KW - PnP Solver
KW - RANSAC
KW - Spacecraft Pose Estimation
KW - Zeroth-Order Optimization
UR - https://www.scopus.com/pages/publications/105044248713
U2 - 10.1109/ICAISISAS68969.2026.11567784
DO - 10.1109/ICAISISAS68969.2026.11567784
M3 - 会议稿件
AN - SCOPUS:105044248713
T3 - 2026 Joint International Conference on Automation-Intelligence-Safety, ICAIS 2026 and International Symposium on Autonomous Systems, ISAS 2026
BT - 2026 Joint International Conference on Automation-Intelligence-Safety, ICAIS 2026 and International Symposium on Autonomous Systems, ISAS 2026
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2026 Joint International Conference on Automation-Intelligence-Safety, ICAIS 2026 and International Symposium on Autonomous Systems, ISAS 2026
Y2 - 8 May 2026 through 10 May 2026
ER -