Skip to content

Benchmarking of Pointnet-Based Architectures for Pose Estimation of Non-Cooperative Spacecraft from LiDAR Data

Jul 2026 · IEEE International Workshop on Metrology for AeroSpace · pp. 58-63 · 0 citations · 25 references

Abstract

In the framework of space missions involving Close-Proximity Operations (CPO), accurate pose estimation between a chaser spacecraft and a non-cooperative resident space object remains a significant challenge. In such scenarios, pose determination must rely solely on Electro-Optical sensors such as cameras or LiDARs. Recently, Deep Learning (DL) approaches have gained significant attention for this task, demonstrating promising performance in terms of accuracy and runtime efficiency. This work presents a comparative analysis of both original and state-of-the-art point-based neural network architectures for LiDAR-based pose initialization of noncooperative spacecraft. The evaluated models, based on PointNet, differ in encoder adopted - PointNet (PN), a two-layer PointNet-based encoder (PN-PCN), and PointNet++ (PN++) - and in attitude parametrization, including quaternion, axisangle, 6D, and soft-label representations. The predicted pose is refined using the Iterative Closest Point (ICP) algorithm. Performance is assessed on synthetic scans of different targets generated by an in-house realistic LiDAR simulator. Results show that the PN-PCN encoder provides the best accuracyspeed trade-off, while soft-label parametrization yields the most accurate attitude estimation. With this configuration, subdegree-level attitude errors are achieved for Envisat and Aura, while degree-level attitude errors are obtained for Apollo, due to its more challenging symmetric geometry. Translational errors remain within a decimeter-level range across all targets.

View source

Similar papers

2026

Vision-Based Relative Pose Estimation of Space Objects for Proximity Operations Using Lab-Validated Deep Learning

A scalable framework for real-time, 6-DOF pose estimation of uncooperative space objects using monocular cameras, addressing dataset scarcity by generating mission-specific synthetic data augmented with style randomization to mitigate domain shift.

Matteo Forasassi · 0 citations
Aug 2026

Binary Networks and Continual Learning for Pose Estimation from a Single Aerial Image

Pose estimation using aerial images captured by Unmanned Aerial Vehicles (UAVs) allows the localisation in GPS-denied scenarios. Several methods based on deep learning approaches with convolutional neural networks (CNN) have become tools for estimating localisation from images. However, building a model that can estimate the pose from a single image needs a large dataset and training time to obtain a result. Besides, the model can be inappropriate in assessing the correct pose in dynamic scenarios with multiple changes. Therefore, we propose a methodology using a binary network with a Continual Learning (CL) strategy to create an estimation model during the same flight mission. Also, we use a submap scheme and multiple models to acquire the UAV’s localisation into different parts of the trajectory. Finally, we use PoseNet, ORB-SLAM2 and single-model for comparison purposes in four scenarios, achieving a percentage error of 14% of the total trajectory and a processing time of 51 ms with our proposed approach.

A. Cabrera-Ponce, L. Rojas-Perez, Manuel Martin-Ortiz et al. · 0 citations
Aug 2026

ObsGuide: A Plug-and-Play Observability-Guided Residual Selection Method for Accurate LiDAR Odometry

Light detection and ranging (LiDAR)-based odometry and mapping is a cornerstone of robotic perception and navigation. Recent work has primarily improved the accuracy of LiDAR odometry either by feeding every available residual into the optimizer or by resorting to multisensor fusion; however, the intrinsic information contained in LiDAR residuals has been little explored from the perspective of nonlinear optimization. To address this, we propose observability-guided residual selection (ObsGuide), a novel plug-and-play residual selection method. Rather than developing a standalone system, ObsGuide is designed as a versatile front-end module that seamlessly integrates into existing LiDAR odometry pipelines. It employs a generalized residual evaluation strategy that ranks residuals based on their observability contribution to the six-degree-of-freedom (6-DoF) pose, explicitly accounting for the planarity or linearity of geometric features. By actively retaining only the minimal subset of residuals that impose the strongest pose constraints during optimization, ObsGuide enables existing pipelines to achieve higher accuracy with significantly fewer residuals. Extensive experiments—conducted by integrating ObsGuide into standard baselines (both optimization- and filter-based) across public benchmarks and real-world indoor and outdoor sequences—confirm its effectiveness, runtime efficiency, and strong generalization ability.

Yufei Lu, Qun Hao, Shaohui Zhang · 0 citations
Preprint Aug 2026

AirAlign: Geometry-Aware Relative Pose Alignment for UAV Last-Meter Navigation

Unmanned aerial vehicle (UAV) navigation in modern low-altitude environments requires more accurate pose alignment in the final approach stage for target information acquisition or manipulation, making"last-meter"navigation increasingly important. However, severe viewpoint and appearance variations make this task challenging. To tackle this problem, we propose AirAlign, a framework for RGB-only image-pair relative pose alignment for UAVs. AirAlign uses a pretrained visual geometry reconstruction model as the backbone to extract geometry-aware features from source-target image pairs. In addition, to better utilize the limited training data, we split the training set into multiple scene-disjoint folds for unseen cross-validation and model selection. During inference, the predictions of the selected models are averaged to form the ensemble output of the overall framework. Experiments on the PairUAV challenge at the ACMMM 2026 Workshop on UAVs in Multimedia demonstrate the effectiveness and robustness of our method, while comprehensive ablation studies validate the contribution of each component.

Jin Zhou, Shuo Feng, Yufei Wu et al. · 0 citations