Multi-spacecraft collaborative attitude determination of space tumbling target with experimental verification

Highlights

• An attitude determination approach based on multi-spacecraft collaborative observation is proposed.

• The presented approach can successfully estimate the attitude of the tumbling target with no prior information.

• Numerical simulations and ground micro-gravity experiments are carried out to demonstrate the proposed method.

Abstract

The precise attitude determination of a tumbling non-cooperative space target is challenging. This paper proposed an attitude determination approach based on multi-spacecraft collaborative observation. The three-dimensional (3-D) point cloud data of the target is constructed in terms of the collaborative observation data. By virtue of the target features and point cloud matching between two adjacent frames, the attitude change information of the target is derived, and based on this, the attitude information of the tumbling target is estimated via Extended Kalman Filter (EKF). Numerical simulations and ground micro-gravity experiments are carried out to demonstrate the performance of the proposed collaborative observation method. The results show that the presented approach can successfully estimate the attitude of the tumbling target with no prior information, and avoid the presence of occlusion or data missing during the observation process.

Introduction

With the increasing prominence of space debris and malfunctioned spacecraft, the demand for on-orbit operations such as active debris removing (ADR), on-orbit refueling has received considerable attention [[1], [2], [3], [4]]. Among these missions, the precise attitude determination of a non-cooperative space target is viewed as the prerequisite for the implementation of on-orbit servicing operations. Unlike the traditional cooperative target, the prior motion and structure information of the non-cooperative target is generally unavailable, which makes the attitude determination of the tumbling target is challenging.

In literature, a number of attitude determination algorithms have been put forward [[5], [6], [7], [8]]. Biondi G proposed a vision-based measurement method for attitude identification [9], in which the target spacecraft's feature points and 3-D geometry are assumed to be known a prior. Based on 3-D point cloud morphing, Li Y proposed a novel framework for estimating the relative position and attitude of a non-cooperative target [10], in which the motion information of the target can be observed continuously and its 3-D point cloud can be measured by stereovision. However, the framework does not consider the situation without enough measurements and relies on continuous observation. In view of the occlusion phenomenon and the observable condition caused by the tumbling feature of the non-cooperative target, the loss of feature points and measurement data is inevitable during the observation process. Based on optical flow method and stereo matching, Oumer NW predicted the attitude of the target relative to the observer in the presence of occlusions [11]. However, the proposed method relied on optical flow technology, which is particularly sensitive to the light. In work [7], based on the assumption that the characteristic information of the target is known as a prior, several feature points of the target are tracked by binocular vision measuring system, and the angular rate of space debris is estimated by combining compressed sensing with Kalman filtering. However, this work assumes that the prior information of the target's characteristic position can be obtained directly. Consider the lack of reliable visual features and the robustness for lighting condition, Sumant Sharma introduced the spacecraft pose network (SPN) for on-board attitude determination by using convolutional neural network (CNN) [12]. Nevertheless, the massive datasets necessary to train machine learning algorithms are typically difficult to obtain in real space operations. More significantly, it should be noted that the above-mentioned works suffer from the limitation of assuming that the model and/or feature points information of the target can be (partial) observed prior, which is generally unavailable when the target is non-cooperative, such as a malfunctioned satellite.

Additionally, the above-mentioned studies mostly focus on single observer, including monocular camera, binocular camera or multi-ocular camera [13,14]. As mentioned previously, the navigation and observation of non-cooperative target based on single observer has obtained many achievements. Even with the monocular camera, the parameters of relative motion can be determined with a sufficiently high accuracy base on angle-only measurements [[15], [16], [17]], and the attitude information can also be estimated with three or more identified feature points on the target. However, due to the unidentified feature points and the features lost during the tracking, the attitude information of tumbling target cannot be resolved by single observer. Since the point cloud of the space target cannot be obtained by monocular camera, the attitude determination based on point cloud matching can also not be applied [18]. On the contrary, deepness images, containing the depth information and 3-D point clouds of target, can be obtained by binocular or multi-ocular camera fixed on single spacecraft. However, the measuring range and accuracy of the stereo vision camera is limited by its baseline length. In addition, the poor space lighting conditions or observation position may result in failure of the attitude determination based on single observer. Given the limited performance of single observer, spacecraft formation has been gradually considered and studied. Felicetti L proposed an in-orbit space debris surveillance algorithm utilizing a network of distributed optical sensors carried onboard multiple spacecraft flying in formation [19]. Yuan J designed a dual vector quaternion-based fault-tolerant pose and inertial parameters estimation algorithm utilizing dual small satellites [20]. However, each vehicle has to estimate the attitude and other information of target in these methods, and the collaboration of multiple vehicles is adopted in order to improve the accuracy and stability.

For this problem, this paper proposes a novel attitude determination approach based on multi-spacecraft collaborative observation. The contributions of this paper are summarized as follows:

• The target's 3-D point cloud data is real-timely reconstructed by collaborative observation of multiple spacecraft with different viewpoints. Compared with the single spacecraft-based observation method, the proposed approach can estimate the attitude information of the tumbling target without prior information of the target, and avoid the occlusion or data missing during the observation process.

• Based on the target features and point cloud matching between two adjacent frames, the attitude change information of the target is derived, and based on this, the attitude information of the tumbling target is estimated via EKF methodology.

• The ground experimental verification of the proposed method is carried in the six-DOF ground neutral buoyancy micro-gravity experimental platform. The result demonstrates the proposed method can successfully estimate the attitude of the tumbling target with no a prior information, and avoid the presence of occlusion or data missing during the observation process.

This paper is organized as follows. Section 1 gives the collaborative observation scheme for non-cooperative target from multiple viewpoints. Section 2 introduces the detailed flows to obtain the attitude change information from the image sequences and the attitude determination of the target via EKF method. Section 3 analyses the influence of phase angle for multi-spacecraft collaborative observation. The experimental verification based on neutral buoyancy micro-gravity platform is performed in Section 4. Section 5 concludes the paper.