Polarity--Time Interaction and Multi-Level Reliability for Event-Based Spacecraft Pose Estimation

Yonglong Zhang, Zongwu Xie, Yang LiuarXiv 2026 · arXiv

TL;DR

This paper introduces a multi-level reliability framework for spacecraft pose estimation using event cameras, which are ideal for the high-contrast, high-speed environment of space. By intelligently weighting the "trustworthiness" of data at the keypoint, frame, and trajectory levels, the authors achieved state-of-the-art results on the SPADES challenge, outperforming existing methods by approximately 5.4%.

What Problem Does This Paper Address?

Estimating the 6-DOF pose (position and orientation) of a non-cooperative spacecraft is a critical task for autonomous docking and satellite servicing. While traditional cameras often fail due to extreme lighting conditions and motion blur, event cameras offer a promising alternative by reporting asynchronous brightness changes with microsecond resolution.

However, event data is sparse and noisy. Keypoint detection—the process of finding specific landmarks on the spacecraft—often results in uneven errors. A "sharp" detection in the image doesn't always translate to a precise 3D pose, especially if the geometry of the spacecraft or the camera's viewpoint is unfavorable. The authors tackle the challenge of how to quantify and propagate this uncertainty through the entire estimation pipeline, from individual pixels to the final flight trajectory.

Methods

The authors propose a three-stage pipeline designed to handle uncertainty at every step:

  1. Reliability-aware Pose Initialization (RPI): The system uses a three-branch neural network to process positive events, negative events, and the "recency" of events (timing). These branches interact via cross-branch attention to build a robust feature representation. Crucially, a "reliability head" predicts the precision of each detected keypoint, weighting them based on both their local appearance and how their error might propagate into the final pose.
  2. Event–CAD Consistency Refinement (ECR): Once an initial pose is found, the system refines it by comparing the event edges to the spacecraft's known CAD model. They use a "direction-reversal-invariant orientation moment" to ensure that the matching is robust, regardless of whether an event was triggered by a light-to-dark or dark-to-light transition.
  3. Multi-source Quality-aware Kinematic Smoothing (MQKS): At the sequence level, the system performs trajectory optimization. Instead of treating every frame as equally important, it uses a 10-dimensional quality descriptor (incorporating heatmap shape, reprojection consistency, and geometric conditioning) to weight each observation's contribution to the final smoothed path.
Fig. 1. Event-based spacecraft pose estimation from detection to trajectory recovery. RTMDet-tiny detects the target, local harmonic fitting smooths the bounding-box trajectory, and the resulting boxes define the five-window ROI sequence. RPI establishes semantic correspondences and recovers an initial pose. ECR refines the pose using event–CAD agreement under geometric safeguards. MQKS combines the resulting observations with time-aligned quality and motion constraints to estimate the trajectory. Solid arrows denote the estimation flow and geometric inputs; the dashed arrow carries auxiliary quality information.

Datasets

The research utilizes the SPADES (Spacecraft Pose Estimation Dataset using Event Sensing) dataset. This benchmark includes:

  • Synthetic Trajectories: 300 trajectories containing nearly 180,000 frames.
  • Laboratory Trajectories: 31 real-world laboratory sequences (15,500 frames) providing a bridge to physical hardware performance.

Key Insights and Ideas

  • Timing is a First-Class Citizen: Simply counting events isn't enough. By including a dedicated branch for "recent-event timing," the network gains a sense of temporal recency that significantly improves localization.
  • Geometry Matters for Reliability: A key "aha moment" is that reliability shouldn't just be based on heatmap confidence. The authors use a pose-propagation proxy that links learned weights to their actual effect on translation and rotation, rather than just how "concentrated" a detection looks in 2D.
  • Multi-Level Safeguards: The ECR stage includes geometric safeguards that prevent the system from accepting a "refined" pose if it doesn't actually improve the consistency between the CAD model and the observed event edges.
Fig. 2. Reliability-aware pose initialization. (A) Positive counts, negative counts, and recent-event timing interact through cross-branch attention and channel gates. The projected interaction residual is added to the polarity representation before five-window STA and spatial feature extraction. (B) Decoded semantic coordinates and local heatmap responses provide geometric and response-shape descriptors for point reliability. Predicted relative precision weights the geometric pose refinement.

Key Results

  • State-of-the-Art Performance: The complete system achieved a combined pose score of 0.0474 on the official SPADES challenge test set, the lowest (best) score among listed methods and 5.4% better than the previous leader, PASAN+IRIS.
  • Ablation Gains: The jump from a base localization system (0.0772) to the full reliability-aware cascade (0.0514 on validation) represents a 33.4% reduction in error.
  • Efficiency: The default five-window system runs at 52.7 ms/frame, making it potentially viable for real-time or near-real-time space applications.
  • Smoothing Impact: Quality-aware weighting in the MQKS stage alone reduced the error by 6.4% compared to standard constant-weight trajectory smoothing.

Limitations and Future Work

While the system is highly effective, the authors noted a persistent observation bias in some continuous sequence examples, where the estimated trajectory, though smooth, remained slightly offset from the ground truth.

Future work could focus on:

  • Reducing this residual bias through better sensor calibration or more sophisticated motion models.
  • Exploring even more compact network architectures for deployment on resource-constrained neuromorphic hardware in actual spacecraft.

Summary updated

Abstract

Sparse event responses produce uneven keypoint errors whose effects on spacecraft pose depend on correspondence geometry and the reliability of successive observations. We present a reliability-aware framework connecting polarity--time localization, geometric recovery, event--CAD refinement, and trajectory optimization. A three-branch representation combines positive events, negative events, and recent-event timing through cross-branch attention and gating, followed by spatiotemporal aggregation. An ordered semantic-point head estimates relative correspondence precision for geometrically constrained pose recovery. Event-edge and CAD-projection agreement then guide local pose refinement under keypoint-geometry safeguards. At the sequence level, a ten-dimensional descriptor combines heatmap shape, reprojection consistency, and geometric conditioning to set time-aligned observation scales. On the official SPADES challenge test set, the complete system achieves a combined pose score of 0.0474, approximately 5.4\% below PASAN+IRIS. Validation ablations reduce the score from 0.0772 for base localization with uniform PnP to 0.0514 for the complete cascade; on identical single-frame observations, quality weighting reduces the score from 0.0549 with constant-weight optimization to 0.0514.
BibTeX
@misc{zhang2026polarity,
  author = {Yonglong Zhang and Zongwu Xie and Yang Liu},
  title = {{Polarity--Time Interaction and Multi-Level Reliability for Event-Based Spacecraft Pose Estimation}},
  year = {2026},
  eprint = {2610.36625},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.36625},
}
Figure