Advances in Space Robots for On-Orbit Servicing: A Comprehensive Review

and operation mode are presented. Moreover, ﬁ ve key technologies of space robots are investigated in detail: visual perception, motion planning and control, multifunctional end-effectors, ground teleoperation, and ground veri ﬁ cation, and the relevant technological challenges have been highlighted. In addition, seven future research prospects are discussed, including advanced control strategies, visual perception and capture technologies, ground teleoperation based on multisensory virtual/ augmented reality, crossover research between space robots and spacecraft, performance analysis with space – ground differences, soft space robots, and recon ﬁ gurable modular space robots. Insummary


Introduction
The exploration and exploitation of space resources have become attractive and feasible human activities with rapid advances in space science and technology. There are many on-orbit servicing (OOS) missions, such as inspecting, capturing, refueling, and repairing satellites, orbital debris removal, and assembling and maintaining large space infrastructure. [1][2][3][4] Some early OOS missions at low Earth orbit (LEO) have been accomplished by astronaut extravehicular activities (EVAs). However, unfavorable features of the space environment, such as a high vacuum, microgravity, intense radiation, and large temperature differences, usually make space operations very risky and complicated for astronauts. Space robots are reliable and robust enough to accomplish various OOS missions that may pose a danger or be inaccessible to astronauts, such as geosynchronous orbit (GEO), and can improve the safety and efficiency of on-orbit operations.
Over the past four decades, space robots for OOS have attracted the interest of many researchers, and some representative engineering applications and technology verifications have been conducted on space shuttles, outside/inside space stations, and on satellites. [5][6][7] The preliminary results reveal that space robots can assist or replace astronauts in completing various missions, such as aiding astronauts in repairing the Hubble Space Telescope, constructing space stations, and maintaining satellites. [8][9][10] Modern space robots featuring high intelligence and performance are required as OOS missions increase in complexity, and corresponding breakthroughs in critical technologies are urgently needed.
This article begins with a review of representative space robotic programs for OOS and the development trends of space robots in Section 2. Subsequently, the primary key technologies and challenges for using space robots are investigated in Section 3, covering aspects such as visual perception, motion planning and control, multifunctional end-effectors, ground teleoperation, and ground verification. Section 4 presents seven prospects for future research on space robots. Finally, Section 5 summarizes the article with concluding remarks.
(DOFs) and a 15.2 m length. It has completed many OOS missions on the space shuttle, including deploying and recovering payloads, constructing the ISS, and supporting astronaut EVAs to repair the Hubble Space Telescope, as shown in Figure 2a. [11][12][13][14][15] The SRMS is the first space manipulator system in human history, providing significant experience in developing the Space Station Remote Manipulator System (SSRMS).
The ROTEX, initiated by the German Aerospace Center (DLR), was launched in April 1993. It has 6 DOFs and is about 1 m long, as depicted in Figure 2b. [16] Four teleoperation modes were verified on the Space Shuttle Columbia: automatic, teleoperation on board, teleoperation from ground, and tele-sensorprogramming. It has performed three basic tasks, including assembling a mechanical truss structure, connecting and disconnecting an electrical plug, and grasping a floating object. [17][18][19] The ROTEX verified the ground teleoperation technology of a space robot in orbit for the first time.
The MFD was developed by the National Space Development Agency of Japan (NASDA), now the Japan Aerospace Exploration Agency (JAXA), and was launched in August 1997. It has 6 DOFs and is about 1.5 m long. Tests such as attachment and detachment of the orbital replacement unit (ORU), as shown in Figure 2c, [20] opening and closing of the hinged door, and verification of ground teleoperation were completed on the Space Shuttle Discovery. [21][22][23] The purpose of the MFD tests was to verify the performance of the small fine arm of the Japanese Experiment Module Remote Manipulator System (JEMRMS) outside the ISS.
In summary, space shuttle-based space robots are the pioneers of space robotics. The SRMS, ROTEX, and MFD have been crucial to engineering applications and key technology verifications.
The SRMS end-effector is specially designed to capture cooperative targets through on-orbit teleoperation by astronauts. Moreover, the ROTEX and MFD verified the technologies for ground teleoperation, and the ROTEX compensated for the impact of the long 5-7 s space-ground communication time delay using predictive graphics simulation. Consequently, these programs laid the foundation for developing space robots outside/inside the ISS.

Space Robots outside/inside the ISS
On the one hand, the construction and maintenance of the ISS rely on space robots, while on the other hand, the ISS provides a valuable platform for essential space robot technology verifications. Representative programs include the Mobile Servicing System (MSS), JEMRMS, Robotics Component Verification on ISS (ROKVISS), European Robotic Arm (ERA), GITAI S1, Robonaut 2 (R2), and Skybot F-850.
The MSS, organized by the Canadian Space Agency (CSA), is mainly composed of the SSRMS, the Special Purpose Dexterous Manipulator (SPDM), and the Mobile Base System (MBS), as shown in Figure 3a. [9,[24][25][26][27] The SSRMS (also known as Canadarm2), launched in April 2001, is a 7-DOF manipulator with a symmetrical structure. [28] It is 17.6 m long, has a maximum payload of 116 000 kg, and can walk around on the ISS, making it significantly more flexible than the SRMS. Multiple tasks such as supporting astronaut EVAs, assembling the ISS, replacing the ORU, and capturing cargo craft, as shown in Figure 3b, [29] were completed in orbit. [30][31][32] The SPDM (also known as Dextre) was launched in March 2008 and is a dualarm robot system with 15 DOFs. It has two 7-DOF manipulators and a body rotary joint. [33] It is about 3.5 m long, with a maximum payload of 600 kg. The SPDM is installed at the end of the SSRMS to complete spacecraft module replacement or the robotic refueling mission (RRM), as shown in Figure 3c, [34] and other delicate operations through an end-effector that can change various operating tools. [35][36][37] The MBS, launched in June 2002, is a support platform and can provide power and data links for SSRMS and SPDM. It can move along the ISS truss, as indicated in Figure 3d. [38] Depending on the cooperative operations of the SSRMS and SPDM, the MSS plays an essential role in ISS construction and maintenance.
The JEMRMS, developed by JAXA, was launched in June 2008 and mounted outside the Japanese Experimental Module of the ISS, as illustrated in Figure 4a. [39,40] It consists of a main arm and a small fine arm connected in series to form a macro-micro manipulator system, both of which are 6-DOF manipulators. The main arm, about 10 m long with a maximum payload of 7000 kg, can operate, position, and handle large payloads. The small fine arm is about 2 m long with a maximum payload of 300 kg and is applied to fine operation tasks. The primary function of the JEMRMS is to maintain the exposed payloads outside the cabin. It successfully completed OOS missions using the adaptive control for vibration suppression of the macro-micro manipulator. [41][42][43] ROKVISS, initiated by DLR, was launched in December 2004 and mounted outside the ISS Russian Service Module in January 2005. It has 2 DOFs, as indicated in Figure 4b. [44,45] During OOS period, it mainly verified the dynamic characteristics of DLR's modular and lightweight joint, as well as different control modes such as automatic control and force feedback teleoperation. ROKVISS reduces the long space-ground time delay to below 20 ms through a dedicated communication link that improves the telepresence of ground teleoperation. [46][47][48] The ERA, designed and assembled for the European Space Agency (ESA) with Airbus Defence and Space Netherlands as the prime contractors, was launched in July 2021. Up to eight European countries have participated in this industrial undertaking. The ERA is a 7-DOF, 11.3 m-long manipulator with a maximum payload of 8000 kg, as shown in Figure 4c. [49] It can walk outside the Russian cabin of the ISS and conduct OOS missions such as installing and replacing ORUs, supporting astronaut EVAs, processing external payloads, and visually inspecting the ISS outer surface. The ERA aims to provide services for the construction and maintenance of the ISS. [50][51][52] GITAI S1, designed by the GITAI company in Japan, was launched in August 2021. It has 8 DOFs and is 1 m long. GITAI S1 executed two tasks autonomously for technology verifications inside the ISS: assembling structures and panels for in-space assembly (ISA) and operating switches and cables for intravehicular activity (IVA), as depicted in Figure 4d. [53][54][55] Figure 2. Space robots on space shuttles for OOS. a) SRMS supporting astronaut EVAs to repair the Hubble Space Telescope. Reproduced with permission. [12] Copyright 2013, NASA. b) ROTEX. Reproduced with permission. [16] Copyright 1993, DLR. c) MFD attaching the ORU. Reproduced with permission. [20] Copyright 1997, JAXA.
These successful on-orbit technical demonstrations have helped develop general-purpose autonomous technologies and accumulated experience of GITAI's future extravehicular space robots.
The humanoid space robot R2 was developed by the U.S. National Aeronautics and Space Administration (NASA) and General Motors and launched in February 2011. Its upper limb has 42 DOFs, including two 7-DOF arms, two 12-DOF hands, a 3-DOF neck, and a 1-DOF waist. Moreover, R2 has more than 350 sensors, characterized by a strong force-sensing capability, a wide range of motion, and high flexibility. During the orbit, tests conducted included human-robot interaction, operating buttons and panels, and capturing floating objects in teleoperation mode. [56][57][58][59] Two 7-DOF humanoid lower limbs were installed on R2 in 2014 to achieve better maneuverability, as revealed in Figure 5a. [60] Currently, R2 is the most intelligent space robot system and the first humanoid robot to serve in orbit.
Skybot F-850, another humanoid space robot, was designed by the Russian Federal Space Agency (RKA) and launched in August 2019. It can autonomously realize human-robot interaction and can also be teleoperated by the operator using a wearable device. Skybot F-850 monitored and reported the spacecraft's flight states during the launch period. While in orbit, Skybot F-850 completed tests such as opening and closing the door and transferring tools, as shown in Figure 5b. [61][62][63] Skybot F-850 is the first space robot that replaces the astronaut and reports the flight states during the launch phase.
In conclusion, space robots and the ISS complement each other. The ISS has been used to verify critical space robot technologies that play an essential role in the construction and maintenance of the ISS. In particular, ROKVISS has verified the dynamic characteristics of space robot joints and force feedback teleoperation with telepresence in orbit. GITAI S1 has demonstrated technology, versatility, dexterity, and safety in microgravity. R2 and Skybot F-850 are highly intelligent humanoid space robots for which human-robot interaction tests have been conducted in orbit. Their humanoid dexterous hands integrated with rich sensors have been used to perform more refined OOS tasks through autonomous control or on-orbit teleoperation. Furthermore, the SSRMS has experienced from on-orbit teleoperation in the early time to the current ground teleoperation. The SSRMS assembled the ISS and captured cargo craft, the SPDM completed the RRM, and the JEMRMS maintained exposed payloads, providing important support to ISS daily operations. Therefore, space robotics technologies are gradually maturing because of the valuable platform that the ISS provides for experiments and applications.

Space Robots outside/inside the CSS
Regarding China's space robots, the Tiangong-2 (TG-2) space robot has conducted several key technology verifications inside the TG-2 Space Laboratory, accumulating valuable technical experience for developing the Chinese Space Station Remote Manipulator System (CSSRMS).
The TG-2 space robot, developed mainly by the Harbin Institute of Technology (HIT), was launched in September 2016. It consists of a 6-DOF manipulator and a five-finger humanoid dexterous hand. The astronaut and space robot coordinate to carry out several on-orbit tests, such as identifying dynamic parameters, grasping floating objects, a handshake between the robot and an astronaut, on-orbit maintenance, and teleoperation in orbit. For simulated maintenance, the robot completed tasks such as tearing multilayer protection, using an electric tool to loosen a bolt, and screwing in an electrical connector, as shown in Figure 6. [6,[64][65][66] The successful implementation of the TG-2 space robot has provided valuable experience for CSSRMS development and application.
The CSSRMS, designed primarily by the China Academy of Space Technology (CAST) and HIT, consists of the Core Module Manipulator (CMM) and Experimental Module Manipulator (EMM), as illustrated in Figure 7a. [67] The CMM and EMM were launched in April 2021 and July 2022, respectively. Both the CMM and EMM are 7-DOF manipulators with symmetrical structures, about 10 and 5 m in length, with maximum payloads of 25 000 and 3000 kg and maximum positioning accuracies of 45 and 10 mm, respectively. [68] They can walk around on the CSS. The CMM is widely used in large-scale transfer tasks because of its long length, and its main OOS missions include module transfer, auxiliary docking, equipment installation, on-orbit maintenance, state monitoring outside the cabin, and supporting astronaut EVAs, as indicated in Figure 7b. [6,69,70] Because of its relatively short length, the EMM enables high positioning accuracy and is applied to certain delicate operations. Its main tasks include payload maintenance, payload handling, state monitoring outside the cabin, and EVA support for astronauts, as shown in Figure 7c. [71,72] Furthermore, the CMM and EMM can work independently or be connected in series to form a macro-micro manipulator system for cooperative work, thereby expanding the working space. As a major landmark mission, the CMM and EMM have cooperated in series for the first time to support EVA for a Shenzhou-14 (SZ-14) astronaut, as highlighted in Figure 7d. [73] Accordingly, the CSSRMS is one of the key facilities for CSS construction, operation, maintenance, and expansion.
The CSS was constructed in orbit by the end of 2022 and then entered its operation stage. Previously, many key technologies have been tested and verified. The TG-2 space robot used a five-finger dexterous hand to perform various operations, verifying the software and hardware design, human-robot cooperation, and on-orbit teleoperation. As the first on-orbit space robot experiments for China's manned spaceflight, the TG-2 space robot provided valuable experience for developing the CSSRMS. For engineering applications, the CSSRMS utilizes two 7-DOF manipulators, the CMM and EMM, to perform a wide range of dexterous operations. [74] Specifically, they walked around on the CSS, docked with the cooperative markers, monitored the states outside the CSS, and supported EVAs for SZ-12, SZ-13, and SZ-14 astronauts. In addition, the CMM transposed the Tianzhou-2 (TZ-2) cargo craft. [75] Therefore, the CSSRMS is a significant piece of equipment for OOS of the CSS.  [67] Copyright 2022, China Media Group, CCTV.com. b) CMM supporting EVA for an SZ-13 astronaut. Reproduced with permission. [69] Copyright 2021, China Media Group, CCTV.com. c) EMM supporting EVA for an SZ-14 astronaut. Reproduced with permission. [71] Copyright 2022, China Media Group, CCTV.com. d) CMM and EMM cooperatively supporting EVA in series for an SZ-14 astronaut. Reproduced with permission. [73] Copyright 2022, China Media Group, CCTV.com. Reproduced with permission. [6] Copyright 2021, editorial office of Chinese Journal of Aeronautics.

Space Robots on Satellites
Space robots on satellites also have significant application prospects, and OOS has been accomplished in several programs, including the Engineering Test Satellite VII (ETS-VII), Orbital Express (OE), Shiyan-7 (SY-7), and Aolong-1 (AL-1) space robots. The ETS-VII was developed by NASDA and launched in November 1997. It is equipped with a 2 m-long, 6-DOF manipulator, as shown in Figure 8a. [76] The ETS-VII utilized virtual graphics prediction and bilateral force feedback technologies to overcome long space-ground time delays. It completed tests such as monitoring and capturing a target satellite, grasping a floating object, installing the ORU, plugging and unplugging an electrical connector, and refueling through on-orbit local autonomous and ground teleoperation. [10,[77][78][79] It verified the Reaction Null-Space method when the satellite's base was floating. [80] The ETS-VII is the first satellite equipped with a manipulator and the first free-flying space robot.
The OE program organized by the U.S. Defense Advanced Research Projects Agency (DARPA) is composed of the Autonomous Space Transport Robotic Operations (ASTRO) satellite and the Next-Generation Satellite (NextSat) and was launched in March 2007. The ASTRO satellite was equipped with a 6-DOF manipulator with a length of 3 m, as depicted in Figure 8b, [81] which was used to conduct on-orbit maintenance on NextSat-simulated failures. Moreover, the manipulator completed OOS tests, including on-orbit capture, refueling, and ORU replacement, promoting rapid technology development for on-orbit autonomous target satellite maintenance. [82][83][84][85] The SY-7 space robot was launched in July 2013, as indicated in Figure 8c. [86] Some key technologies of space robots such as on-orbit maintenance have been verified. [87,88] SY-7's successful experiments have enabled China to progress significantly in critical OOS technologies.
The AL-1 space robot with 6 DOFs initiated by the China Academy of Launch Vehicle Technology (CALT) was launched in June 2016, as indicated in Figure 8d. [89] OOS tests such as capturing space noncooperative simulation debris and maintenance have been conducted. The AL-1 verified some key technologies such as actively removing space debris, and obtained important OOS data. [89,90] In summary, these programs have conducted several technology verifications of on-orbit maintenance. The ETS-VII used virtual graphics prediction and bilateral force feedback to overcome the impact of long space-ground time delays for ground teleoperation. The OE is more intelligent and can autonomously approach, dock, capture, and repair a target satellite. Moreover, China's satellite-based SY-7 and AL-1 space robots have demonstrated on-orbit maintenance technologies and have accumulated valuable experience from OOS experiments. In addition, it is worth noting that a satellite-based space robot is a free-floating system. The satellite's jet device and flywheel or the manipulator's trajectory planning can compensate for the base disturbances, and the ETS-VII used the Reaction Null-Space method to minimize the manipulator's impact force on the base. In the future, space robots are expected to be widely used to service failed satellites to extend their life in orbit.

Space Robot Development Trends
The details of advances in representative space robotic programs are summarized in Table 1, including the country, launch time, application scenario, purpose, and major OOS missions.
Furthermore, the increasing complexity of OOS missions using space robots has resulted in the following development trends for configuration, mission type, target type, and operation mode: Development from nonredundant configurations to dexterous configurations. The early-launched space shuttle-based SRMS, ROTEX, and MFD are 6-DOF nonredundant configurations, while the SSRMS, CMM, EMM, and ERA are 7-DOF dexterous configurations, and GITAI S1 has 8 DOFs. The SSRMS and SPDM, CMM and EMM, and the main arm and small fine arm of the JEMRMS are connected in series to form a macromicro manipulator system. Moreover, the SPDM is a dual-arm system, while R2 and Skybot F-850 are humanoid robots with multiple DOFs. A dexterous configuration is more conducive to enabling space robots to accomplish complex OOS missions. Figure 8. Space robots on satellites. a) ETS-VII system. Reproduced with permission. [76] Copyright 2012, ESA. b) OE system. Reproduced with permission. [81] Copyright 2008, SPIE. c) SY-7 manipulator. Reproduced with permission. [86] Copyright 2006, HIT. d) AL-1 manipulator. Reproduced with permission. [89] Copyright 2017, CALT.
www.advancedsciencenews.com www.advintellsyst.com Development from simple assembly to complex maintenance. The SSRMS conducted simple OOS, such as assembling the ISS and capturing spacecraft. In contrast, the OE executed complex OOS, such as autonomous maintenance and refueling. GITAI S1 performed OOS autonomously for complicated ISA and IVA, and the SPDM completed complex RRM by exchanging four special tools with dual arms. The dexterous robotic hands of R2, TG-2, and Skybot F-850 directly used tools to complete their tasks. In addition, the control mode of space robots has transitioned from simple position-pose control to compliance control with complicated physical contact for the high demands of OOS missions.
Development from cooperative targets to noncooperative targets. On the one hand, the targets on the ISS, CSS, and most satellites usually carry visual markers and interfaces, which are elaborately designed to enable capture by space robots. On the other hand, the end-effectors of most robots, such as the SSRMS, ERA, CMM, and EMM, are specially designed to capture cooperative targets. Moreover, the targets for R2, Skybot F-850, and TG-2 are noncooperative and have been grasped by dexterous robotic hands.
Development from on-orbit teleoperation by astronauts to ground teleoperation by mission experts. Early SRMS and SSRMS operations adopted on-orbit teleoperation by astronauts. In contrast, the ROKVISS, SPDM, CMM, EMM, and the current SSRMS use ground teleoperation by mission experts for most OOS missions, significantly reducing the workload of astronauts and improving the efficiency of space exploration and exploitation.

Key Technologies and Challenges
Space robots execute complicated OOS missions with high safety and reliability in the harsh space environment, which includes zero gravity, complex illumination, and long communication delays. The main phases that should be considered for a typical OOS mission include recognizing, approaching, and capturing the target, as well as the operation mode and ground verification. Therefore, space robots involve many key technologies, including visual perception for noncooperative targets, motion planning and control with a free-floating base and flexibility, www.advancedsciencenews.com www.advintellsyst.com multifunctional end-effectors, ground teleoperation with long time delays, and high-fidelity ground verification. A graphical representation of the primary key technologies is highlighted in Figure 9.

Visual Perception for Noncooperative Targets
A space robot autonomously capturing targets for OOS must be integrated with a visual system that performs visual recognition and pose measurements between the end of the robot and the visual target in real time. These technologies can be divided into cooperative visual perception and noncooperative visual perception according to the characteristics of visual targets. [91][92][93] Cooperative visual perception technologies are currently relatively mature and have been successfully used in many space robots. Some cooperative visual markers are shown in Figure 10. [94,95] However, as for most of actual OOS for faulty satellites, artificial markers are not installed on targets. Therefore, visual perception for noncooperative targets is required but is more difficult because of the unknown target characteristics. The typical features of a noncooperative target are the satellite's adapter ring, solar panel boom, and satellite nozzle, and their characteristic shapes are ellipse, triangle, and cone, respectively, as indicated in Figure 11. The most commonly used perception feature is the adapter ring, and the visual perception includes visual detection and measurement.
Except for orthographic projection, the projection of the adapter ring in an image is a quadratic elliptic curve. The adapter ring's visual detection is essentially the detection of an ellipse. Optimization-based ellipse detection transforms it into a quadratic curve-fitting optimization. Because the algorithm requires constant iteration, its real-time performance is relatively poor, and its ability to detect multiple ellipses is insufficient. [96] Hough Transform-based ellipse detection votes on candidate ellipses in 5D space and extracts the ellipse's contour curves. However, because continuity between the edge points is not considered, it requires significant computation time. [97] Ellipticalarc-features-based ellipse detection method detects the ellipse using the geometric characteristics of the edge curves. The fundamental characteristic element is the elliptical arc rather than a traditional edge point. The ellipse parameters can be obtained by analyzing different elliptical arc combinations, which eliminates  www.advancedsciencenews.com www.advintellsyst.com interference from irrelevant curves, significantly improving the detection speed. This method has received extensive research interest in recent years. [98,99] Visual measurement based on a monocular camera usually depends on some prior information about the target satellite. Because of a lack of depth information, visual measurement cannot be completed independently. [100] Binocular camera-based visual measurement can improve measurement accuracy with more constraint information. Binocular cameras can also back up each other to enhance the system's reliability. However, the stereo matching of the algorithm is time-consuming, resulting in poor real-time performance. [101] Accuracy is improved, and the effect of the change in space illumination is reduced using visual measurement based on multisensor fusion. However, higher power consumption is required, the resolution may be insufficient, and image registration is necessary. [102] Modelbased visual measurement matches the local or overall features of the target satellite in the image to a known prior model. The target's overall visual features are utilized with higher measurement accuracy and stronger anti-interference. However, the algorithm is more complex, requiring relatively significant calculations that cannot meet real-time requirements. [103] Closed-form solutions for the pose parameters are derived by establishing the relationship between the elliptical projection curve and the space circle using constructed-cone-based visual measurement. This approach is simple, and its real-time performance is good, but pose ambiguity cannot be neglected. [104,105] With the gradually increasing number of spacecraft, there will be many more noncooperative targets in future space exploration. Visual perception of space robots is expected to face increasing technological challenges, including highly robust target detection in complex lighting environments and highprecision measurement based on full or partial target model constraints.

Motion Planning and Control with a Free-Floating Base and Flexibility
A space robot must first approach targets to perform OOS missions. Because space robots operate in a zero-gravity environment, the kinematics and dynamics of the manipulator and base are coupled, so approach motion planning must be well designed. Moreover, because gravity loading is not considered, space robots are designed to be lightweight and have a large workspace. They usually have flexible components such as harmonic reducers, joint torque sensors, and long links, which increase flexibility and cause unexpected vibrations that impede precise positioning.
Space robot motion planning can be divided into basecontrolled and base-floating modes according to the characteristics and control strategy of the base. The base-controlled mode case is when the space robot is on a space shuttle or space station, so the robot's mass is far less than the mass of the base, and the base can be regarded as fixed. Therefore, the motion planning for a space robot in base-controlled mode is similar to a ground robot.
Base-floating mode includes free-flying and free-floating modes. [106] In free-flying mode, the thrusters and reaction wheels stabilize the satellite's pose, and its position is not controlled. This mode is usually used in the final approach to the target to ensure that the target is within the robot's workspace. In contrast, all spacecraft thrusters are turned off in free-floating mode, and the position and pose of the base are not controlled, so the spacecraft translates and rotates in response to manipulator motions. This mode is used in the grasping phase because it eliminates sudden motions due to thrusters, conserving propellant and power.
Various optimization indexes, such as minimum energy consumption, obstacle avoidance, and singularity avoidance, should be considered in motion planning to achieve the best space manipulator performance for OOS missions. [106][107][108][109] In addition, the reduction of base disturbances is an important issue for a free-floating space robot, in which kinematic and dynamic coupling exists between the manipulator and the base. The manipulator's motion changes the base's pose, as shown in Figure 12. Therefore, space robot motion planning must also keep the base stable by optimizing the force, controlling the pose, and using other base constraints.
Several effective motion planning approaches have been proposed for free-floating space robots. The generalized Jacobian matrix method constructs the relationship between the motion of the end-effector and nonholonomic dynamic characteristics. [108,109] The motion of the space robot can be planned to reduce dynamic disturbances to the base using the enhanced Figure 11. Typical features of a noncooperative target. www.advancedsciencenews.com www.advintellsyst.com disturbance map approach. [110] In the bidirectional method, both the robot's configuration and the base's pose are controlled simultaneously by calculating the actual robot's forward motion and the virtual robot's reverse motion, as revealed in Figure 13. [111,112] The Reaction Null-Space method was proposed to plan the motion of a space robot with minimal impact force to the base and has been demonstrated by the ETS-VII. [80,113,114] While a space robot is a typical nonlinear rigid-flexible coupling system, the flexible vibration phenomenon must be suppressed to achieve high-precision control performance. The foundation of vibration suppression control is simulating the rigid-flexible coupling dynamics by modeling the flexible joints and links. The flexible joints are usually modeled as linear torsional springs, as depicted in Figure 14, and the friction, nonlinear stiffness, kinematic error, and other factors are considered further. [115,116] The flexible links are generally discretized into systems with a finite number of DOFs. The approaches mainly include the finite element method, [117] assumed mode method, [118] finite segment method, [119] and lumped mass method. [120] The finite element method discretizes the flexible links into finite elements, in which the displacements of the element nodes express the flexible link deformations. The assumed mode method truncates the high-order modes of the flexible links, reducing the equation scale and using a finite number of known modal functions to determine the system's motion. The flexible links are discretized into several rigid segments connected by springs and dampers in the finite segment method to simulate their inertial characteristics. The lumped mass method distributes the mass of the flexible links to several discrete points, which are connected by elastic elements, but its calculation accuracy is low. Furthermore, the primary methods for dynamic rigid-flexible coupling modeling include the Newton-Euler method, [121] Lagrange method, [122] Kane method, [123] and Hamilton's principle method. [124] Vibration suppression control of the space robot can be divided into passive and active control methods. Passive control suppresses the vibrations generated by the flexible components through vibration-absorbing and vibration-isolating structures. However, passive control is relatively inflexible, has limited control capabilities, and the space manipulator's structural characteristics make it challenging to increase structural damping. Thus, active vibration suppression control is widely utilized, with flexibility, robustness, and straightforward engineering application advantages. Active control is divided into model-based and model-free control methods. [125,126] In model-based vibration suppression control, the controller is designed based on a rigid-flexible coupling dynamics model of the space robot. However, the controller design is more complicated when the model parameters are inaccurate or changing because of the unique space environment. Input shaping control forms a new control signal through the convolution of the input signal and the shaping pulse to eliminate the residual vibrations. [127] Adaptive control lightens the disturbance effects from uncertain factors through an online modification of control parameters. [128] Optimal control obtains a control strategy that can improve the space robot's performance for given constraint conditions. [129] Model-free control usually uses closed-loop feedback with the sensors to achieve vibration suppression and can address the problem of inaccurate space robot model parameters caused by uncertain friction and payload changes. Its principle is simple and easily implemented, but the input delay in the feedback loop cannot be ignored. PID control has been widely combined with feed-forward control and other methods to decrease vibrations. [130] The complex high-order system of the space robot is decomposed into two low-order subsystems by singular perturbation control, in which controllers are designed separately according to singular perturbation theory. [131] Sliding mode control adjusts the controller dynamically based on the system states with high response speed and strong robustness. [132] A free-floating base and flexibility are unique characteristics of space robots. The main factors for high-precision motion and control of space robots are suppressing base disturbances and flexible vibrations. However, these have some technological challenges, such as capturing a tumbling target with little disturbance, dynamic parameter identification for high-precision control, and multi-arm coordination for capture and operation considering base disturbances and flexibility.

Multifunctional End-Effectors
An end-effector is mounted on the end of a space manipulator and used to capture a target or complete delicate OOS missions. It can also accommodate the vibrations and endposition errors generated by the space robot's flexible components. Its features include large misalignment tolerance, soft capture, strong load capacity, high rigidity, high precision, high reliability, and intelligent multisensor information fusion. [133,134] In accordance with their various functions, end-effectors can be divided into dedicated and universal end-effector types. Dedicated end-effectors are used for a single target or special tasks, while universal end-effectors are used for dexterous and complicated tasks. Dedicated end-effectors have been widely used in OOS missions because of their excellent capturing ability.
The end-effector of the SSRMS is a three-steel cable-snared mechanism that can be dragged and tightened by three ball screws and locked by four locking mechanisms to achieve mechanical and electrical connections, as indicated in Figure 15a. [135] Its characteristics include large misalignment tolerance, weak collision, and the ability to assemble the ISS and capture moving targets such as cargo craft. [32,136] The SPDM end-effector can change operating tools to perform multifunctional manipulation with high precision. A wire cutter, safety cap, extravehicular robotic nozzle, and multifunction tools were utilized to complete the RRM. [137,138] Moreover, the multifunction tool provides a common interface with four adapters, namely, the tertiary cap, T-valve, ambient cap, and plug manipulator adapters, which are used to accomplish four separate tasks, as shown in Figure 15b. [37] The EMM end-effector is a threefinger capture and locking mechanism evenly distributed on its circumference, with pose tolerance, preloading, grasping, and other fine functions, as presented in Figure 15c, [133,139,140] that can handle and maintain payloads. [72] R2's dual five-finger humanoid dexterous hands are universal with 12 DOFs that can capture general objects and use tools to perform tasks, mainly including operating the task board and using an airflow meter, as shown in Figure 15d. [58,141] The end-effector is essential for facilitating space robot capture and maintenance tasks. With the increasing complexity of OOS missions, end-effectors still have some technological challenges. These will require research on universal end-effectors that can capture noncooperative targets, multifunctional end-effectors that can quickly change tools, and innovative end-effectors to improve performance for large misalignment tolerance and soft capture. Figure 15. Various space robot end-effectors for OOS. a) End-effector of the SSRMS. Reproduced with permission. [135] Copyright 2018, Cision US Inc. b) SPDM multifunction tool with four adapters for the RRM. Reproduced with permission. [37] Copyright 2011, AIAA. c) End-effector of the EMM. Reproduced with permission. [139] Copyright 2022, Official Website of China Manned Space. d) R2 using an airflow meter with dexterous robotic hands. Reproduced with permission. [58] Copyright 2013, AIAA.

Ground Teleoperation with Long Time Delays
Because of the unstructured space environment and complicated OOS missions, human-in-the-loop teleoperation is generally utilized for OOS, combining the intelligence of humans and robots. Astronauts previously operated space robots for manned space missions, and the on-orbit workstation of the CSSRMS is depicted in Figure 16a. [142] Although the on-orbit teleoperation time delay is relatively small, complicated OOS missions are generally time-and energy-consuming, increasing astronauts' workload. ROKVISS established a dedicated communication link with a space-ground time delay of less than 20 ms to verify the telepresence of ground teleoperation. Contour following and spring pulling tasks were completed, and its teleoperation system is illustrated in Figure 16b. [47,143] The communication link limited the experiment time to windows of only 7 min or less as the ISS passed over the tracking station in Germany. [48] Actual OOS missions for engineering applications usually take a long time, so the ROKVISS communication link is only suitable for short-time tests. Previously, ISS astronauts operated the SSRMS to capture cargo craft, but since 2005, more SSRMS operations have been switched to ground teleoperation to reduce the crew's workload. [144] The CSSRMS has been operated through ground teleoperation for most OOS missions, with Figure 16a showing an astronaut monitoring the states in the CSS during a CMMsupported EVA for an SZ-12 astronaut. [142] However, long space-ground time delays are a challenging issue for the continuous ground teleoperation of space robots. The round-trip delay in Earth orbit is between 3 and 10 s, which increases the operator's response time, significantly impacting the stability and transparency of the space-ground control loop. Many approaches have been proposed to deal with these long time delays to improve the telepresence of ground mission experts. These include teleprogramming, bilateral force feedback, virtual graphics prediction simulation, virtual reality, and virtual fixtures. [145,146] Teleprogramming utilizes ground operators to program and simulate the space robot through a human-robot interaction interface, sending verified preprogrammed instruction sequences and forming a remote closed control loop in orbit to execute instructions. This excludes the operator and the time delay from the control loop. [147] Bilateral force feedback overcomes the time delay using a control algorithm without the need to structure the actual working environment. The contact force between the space robot and the environment reacts to the operator, so human intelligence can be added to the control loop. Therefore, reliability and safety can be ensured because the force and position of the ground operator and the space robot are consistent. [148] Virtual graphics prediction simulation creates models Figure 16. Typical teleoperation systems for space robots. a) On-orbit workstation of the CSSRMS. Reproduced with permission. [142] Copyright 2021, China Media Group, CCTV.com. b) Illustration of the ROKVISS teleoperation system. Reproduced with permission. [143] Copyright 2006, ESA. c) Illustration of the ROTEX teleoperation system. Reproduced with permission. [152] Copyright 1993, DLR. d) Illustration of the KONTUR-2 teleoperation system. Reproduced with permission. [157] Copyright 2016, IEEE.
www.advancedsciencenews.com www.advintellsyst.com of the space robot and the actual working environment to predict the states of the system after the time delay based on current states and control inputs. The operator performs ground teleoperation according to the simulation environment to exclude the time delay from the control loop. In addition, the model parameters must be updated during the simulation process to ensure that they are close to the actual environment, making the graphics prediction more accurate. [149] Virtual graphics prediction simulation can also be combined with virtual reality. A virtual reality environment consistent with the actual working environment is reconstructed on the ground using predictive information and information from multiple sensors fed back by the space robot. The ground operator provides control instructions through human-robot interaction equipment and uses visual, tactile, force, and other information to further enhance the telepresence. [150] A virtual fixture can assist the operator in performing operations quickly and accurately in the virtual reality environment. The virtual guidance fixture makes the robot move along a specific path, and the virtual forbidden-region fixture limits the robot's motion area. The accuracy and stability of the operations can be effectively improved through auxiliary guidance by virtual fixtures. [151] The first ground-teleoperated space robot, ROTEX, was implemented in 1993. It adopted tele-sensor-programming, which utilized a predictive graphics simulation and multisensor local autonomy to overcome the long 5-7 s communication time delay, as illustrated in Figure 16c. [17,152] Tests that included assembling a mechanical truss structure, connecting and disconnecting an electrical plugin, and grasping a floating object were completed successfully in orbit. [18] Ground bilateral teleoperation of the ETS-VII manipulator by direct bilateral coupling was conducted to overcome the impact of a 7 s delay and completed slope-tracing and peg-in-hole tasks. The experimental results showed that kinesthetic force feedback to the operator was helpful even under such a long delay, improving the task performance. [10,78,[153][154][155] For the SSRMS ground control, preprogrammed automatic control sequences are utilized to overcome the delay. In this case, the operator is not part of the control loop and is not required to provide continuous inputs to the SSRMS. During operations, the camera views between the on-orbit and ground graphics are compared for verification, as revealed in Figure 17. [144,156] The KONTUR-2 mission aimed to achieve planetary exploration missions, allowing astronauts to work with robots on the ground from an orbital station. During the mission, an astronaut in the ISS used a force-feedback joystick to teleoperate a robot manipulator on the ground. The KONTUR-2 time-domain passivity control approach was employed to handle communication delay, jitter, and data losses. Its teleoperation system is illustrated in Figure 16d. [157] Future OOS missions are expected to utilize ground teleoperation to reduce the workload of astronauts. Limited by long space-ground time delays, there are some technological challenges to improving the telepresence of ground mission experts, such as multimodal telepresence through 3D scene reconstruction and multisensory virtual/augmented reality, bilateral control with variable time delays, and local space robot autonomy.

High-Fidelity Ground Verification
Sufficient high-fidelity ground verification must be performed before launch to guarantee that a space robot operates successfully in space. Typical ground verification emulates microgravity conditions to develop and verify space robot designs. There are six primary experimental systems for ground verification of a microgravity environment: air-bearing, neutral buoyancy, freefall, airplane parabolic flight, suspension, and hardware-in-theloop systems. [3,106] The air-bearing system is widely used to emulate zero gravity in 2D space and includes one rotational and two translational DOFs. This system forms air films between air bearings and a smooth platform, counteracting the robot's gravity and eliminating friction. [8,158,159] Although the experiment time is not limited, microgravity experiments can only be performed in 2D space. The air-bearing system is a fundamental experimental facility for space robot motion tests, such as the CMM's ground experimental system in Figure 18a. [160] The neutral buoyancy system uses a water pool to compensate for gravity in 3D space without time constraints. However, there is water resistance that does not exist in the space environment, and a special sealing design is needed to make the robot waterproof. [161,162] The neutral buoyancy system of the dexterous robot named Ranger is shown in Figure 18b. [161] The free-fall motion of the space robot in 3D space is generated by a microgravity tower in the free-fall system. The experiment time is very short (less than 10 s), the robot's size cannot be too large, and special attention is required to address safety issues. [163,164] The airplane parabolic flight system creates a microgravity environment in a weightless airplane flying along a parabolic trajectory. The experiment is in 3D space, but its duration is also very short (10-30 s). Moreover, the robot's size is limited by the size of the airplane, and the airplane's motion may result in additional vibrations. [165,166] Compensating forces with the same amplitude but in a direction opposite to the tested robot's gravity force can be generated by the suspension system through a cable from an electromechanical system. The experiment can be performed in 3D space, but it only balances the gravity statically, which cannot accurately simulate low-gravity space robot dynamics. In addition, the robot and the suspension system may generate coupled vibrations, and Figure 17. Video survey comparing on-orbit video camera views with graphical models. a) On-orbit video camera views. b) Ground simulation graphics. Reproduced with permission. [144] Copyright 2007, International Academy of Astronautics.
www.advancedsciencenews.com www.advintellsyst.com extra tension is applied to the robot when the suspension cables are not perfectly vertical, which can significantly affect the experimental results. [167,168] The hardware-in-the-loop system is a powerful approach to simulating the complex dynamic behavior of space robots in a space environment, such as the approach, capture, and docking phases. A high-bandwidth hardware robotic system is used to track a trajectory generated by software that simulates the dynamics of the space robot. Drawbacks include a possibly inaccurate model of the space robot and an inevitable time delay between the hardware and software. [169,170] The SPDM task verification facility (STVF) was established to simulate and verify the dynamic behavior of the SPDM while performing ISS maintenance, as shown in Figure 18c. [36,171,172] In addition, a mainstream hardware-in-the-loop system is presented in Figure 18d for OOS missions of a satellite-based space robot. [173][174][175] Space robots and their associated motion planning and control strategies must pass all ground experiments before performing OOS tasks. Therefore, high-fidelity ground verification for microgravity simulation is required to guarantee the successful completion of missions. Nevertheless, the various ground verification systems have differing characteristics. The future technological challenges are to combine multiple experimental systems for relevant ground verifications according to their advantages and to develop innovative ground verification systems to simulate the 3D motion of space robots in a zero-gravity environment with high fidelity and low cost.

Prospects for Future Research
Space robot technologies must advance to accommodate the increasing complexity of OOS missions. After reviewing representative space robotic programs and analyzing the primary key technologies and challenges, the following seven topics are recommended for further space robot research.

Advanced Control Strategies for Nonlinear Rigid-Flexible Coupling Systems
Space robots are time-varying nonlinear rigid-flexible coupling systems. Their dynamic characteristics are complex, with Figure 18. Typical ground experimental systems for space robots. a) CMM's air-bearing system. Reproduced with permission. [160] Copyright 2021, CAST. b) Ranger's neutral buoyancy system. Reproduced with permission. [161] Copyright 2003, Space Systems Laboratory/University of Maryland. c) STVF. Reproduced with permission. [171] Copyright 2004, Wiley Periodicals, Inc. d) Hardware-in-the-loop system for a satellite-based space robot. Reproduced with permission. [173] Copyright 2016, DLR.
www.advancedsciencenews.com www.advintellsyst.com unknown disturbances and uncertainties, and are challenging to model. Strong coupling between their control and dynamic characteristics affects the space robot's performance. For future OOS missions, space robots require a broader operation range, increased load, and higher accuracy, which add additional control requirements. Therefore, the rigid-flexible coupling characteristics must be studied further, and advanced nonlinear control strategies should be developed to realize precise operations. Moreover, the machine learning approach has strong model approximation ability and high adaptability. Machine learning also does not rely on the dynamic model when applied to robot control and can realize autonomous intelligent control. Therefore, machine learning has great potential in the control of space robots.

Advanced Visual Perception and Capture Technologies for Noncooperative Targets
So far, most space robots have been utilized to service cooperative targets using cooperative visual markers and dedicated endeffectors, so the target spacecraft must be equipped with specially designed structures. However, many noncooperative targets, like failed satellites, still need to be serviced by space robots. Therefore, advanced visual perception and capture technologies for noncooperative targets require breakthroughs, mainly in recognizing and extracting target features in complex lighting environments and designing universal end-effectors to accomplish a broader range of OOS missions.

Advanced Ground Teleoperation based on Multisensory Virtual/Augmented Reality
Since space environments are difficult to predict and cannot be modeled in advance, human participation is usually required to make decisions during ground teleoperation. Poor telepresence severely affects the stability of the control system. Therefore, further research on ground teleoperation technologies based on multisensory virtual/augmented reality combining virtual graphics prediction simulation and 3D reconstruction is needed. Such research would improve the operator's telepresence using visual, tactile, force, and other information. Moreover, accurate kinematic and dynamic space robot parameters should be obtained by on-orbit parameter identification to improve model prediction and control accuracy for ground teleoperation. The model parameters of the ground teleoperation system can be updated to realize efficient and delicate OOS operations.

Crossover Research between Maintenance by Space Robots and Maintainability Design of Spacecraft
On-orbit maintenance by space robots and the maintainability design of spacecraft are two crucial aspects of OOS. However, spacecraft are rarely maintained in orbit and can only be decommissioned when running out of fuel or breaking down. Such phenomena have limited long-term space robot development, inhibiting the utilization of the advanced achievements of on-orbit maintenance technologies. Consequently, crossover research between on-orbit maintenance by space robots and the maintainability design of spacecraft should be conducted to enhance spacecraft autonomous maintenance capability and prolong on-orbit work life.

Performance Analysis with Space-Ground Differences
Ground experiments have difficulty accurately imitating the actual on-orbit characteristics of space robots, such as flexible joints and links, the free-floating base, macro-micro manipulator systems, and space environments, which show the important differences between space and ground operations. Thus, the valuable experimental data from space robots in orbit are required to be fully utilized, and the key performance differences between space and the ground need to be analyzed. Subsequently, novel guidelines can be proposed for designing, planning, and controlling future space robots to continuously advance their development.

Advanced Soft Space Robots
Compared with rigid robots, soft robots have more DOFs, and greater dexterity, deformability, and robustness, exhibiting better adaptability to unstructured, narrow, and complex space environments. [176][177][178] In addition, capturing is a typical OOS mission for space robots. A rigid end-effector has a large impact force, while a deformable universal soft gripper can absorb most of the energy generated by collision. A universal soft gripper has large misalignment tolerance, soft capture, and passive compliance performance that are superior to a rigid end-effector so that it can perform OOS missions dexterously. [179][180][181] Therefore, soft robots have broad application prospects for future OOS.

Advanced Reconfigurable Modular Space Robots
Previous space robots have had specific structures, functions, and motion forms, and their ability to perform OOS missions in unknown and complex environments is limited. In recent years, reconfigurable modular robots have been widely studied because of their unstructured characteristics. [182][183][184] They can change the structures for varying OOS missions and environments, forming a range of complex configurations. Modular robots have the advantages of strong adaptability to the environment, low manufacturing cost, self-repair, self-deformation, and versatility. They can execute highly flexible tasks in dynamic, unknown, and unstructured environments. In conclusion, reconfigurable modular space robots have excellent potential for future space exploration.

Conclusion
This article comprehensively summarizes space robot developments, trends, key technologies, challenges, and research prospects. First, the advances in representative space robotic programs for OOS around the world through 2022 are reviewed, and development trends for configuration, mission type, target type, and operation mode are presented. Moreover, five key technologies of space robots are investigated in detail: visual perception, motion planning and control, multifunctional end-effectors, ground teleoperation, and ground verification, and the relevant technological challenges have been highlighted. In addition, seven future research prospects are discussed, including advanced control strategies, visual perception and capture technologies, ground teleoperation based on multisensory virtual/ augmented reality, crossover research between space robots and spacecraft, performance analysis with space-ground differences, soft space robots, and reconfigurable modular space robots.
In summary, with further developments in space robotics, future space robots will be more intelligent, replacing humans to accomplish various complex and dangerous OOS missions efficiently. Space robots are critical equipment that can fulfill human dreams for future space exploration and exploitation. www.advancedsciencenews.com www.advintellsyst.com