ETS-VII

Terrestrelasso, via Wikimedia Commons. CC BY-SA.

The problem

A manipulator on a free-flying satellite has to do two things a fixed-base arm does not. It has to move without upsetting the attitude of the vehicle that carries it, because arm motion exchanges momentum with the base, and it has to be commanded across a ground link whose delay and short passes limit how much an operator can do. Engineering Test Satellite VII was the Japanese mission built to test both, together with autonomous rendezvous and docking between two unmanned spacecraft. It launched on 28 November 1997 [2]. The program is described in full by its project lead in [4]. The chaser, Hikoboshi, carried a 2 m six-degree-of-freedom manipulator; the smaller target, Orihime, was released and recaptured [2].

The arm masses about 140 kg on a 2550 kg base [1], so arm motion produces measurable base attitude change and the attitude controller’s response feeds back into the arm. The two systems are coupled through the base and neither can be designed in isolation. Gravity gradient torque is not negligible on this scale: in free drift the base attitude drifts more than one degree over several minutes [1], and beyond one degree of pointing error the high-bandwidth link is lost, which bounds how long the vehicle can be left in free-floating mode during an arm operation. The ground side has its own limit: round-trip delay from the NASDA ground station through TDRS was 6 to 7 seconds [3], and work is bounded by pass length, not by arm speed. Three successive flight passes were used for the experiment set of 30 September 1999 [1], and the antenna experiment window was about 20 minutes per pass [7].

What has flown

ETS-VII

The mission ran from launch on 28 November 1997 to December 1999, against a planned end of May 1999 [1]. Surveys of space robotics treat it as the first free-flyer arm in orbit and the historical reference for free-floating manipulator dynamics [24].

Parameter	Value	Source
Chaser mass	2550 kg	[1]
Chaser inertia, Ixx / Iyy / Izz	6200 / 3540 / 7090 kg m2	[1]
Target mass	500 kg	[1]
Manipulator	2 m, 6 DOF	[1]
Arm mass	about 140 kg over six links	[1], [5]
Arm attachment point in base frame	(-0.79, -0.29, 1.00) m	[1]
Launch	28 November 1997, H-II Launch Vehicle 6F from Tanegashima Space Center	[2], [15]
International designation	Hikoboshi 1997-074B, Orihime 1997-074E	[15]
Combined launch weight	2,860 kg	[15]
Orbit	circular, 550 km altitude, 35 degrees inclination, period about 96 minutes	[1], [15]
Design life	1.5 years	[15]
Rendezvous-docking experiments conducted	3	[15]
Ground link	NASDA ground stations via the TDRS data relay satellite	[15]

The arm carries a multisensory gripper and an advanced robotic hand supplied by MITI [1], [5]. CCD cameras are carried at the shoulder and on the hand, and the hand carries force-torque sensing used for compliance and impedance control during contact tasks [6]. ETS-VII carried no joint torque sensors, so the mass and inertia of the base and of each link were identified from reaction wheel momentum and rate telemetry in existing flight data, not from a dedicated maneuver [14]. Link parameters are published in full [1]:

Link	Length (m)	Mass (kg)	Izz (kg m2)
1	0.35 (offset 0.28)	35.0	1.69
2	0.87	22.5	3.75
3	0.63	21.9	2.53
4	not stated	16.5	0.072
5	not stated	26.0	0.13
6 (with end effector)	0.53	18.5	0.26

Rendezvous and docking. Sensing is staged by range: GPS relative navigation from 500 m and beyond, rendezvous laser radar from 500 m to 2 m, and a proximity sensor from 2 m [2]. The vehicle holds at a 2 m virtual point for 15 minutes before the docking phase. Docking approach velocity is 1 cm/s, contact acceleration peaks at 14 mm/s2, and the low impact specification is below 1.5 mg, with measured control accuracy of 7 mm lateral against a 25 mm requirement and 0.2 degrees attitude against 1.9 degrees [2].

Three flight programs were run: FP1 on 7 July 1998 separated at 1.8 cm/s and redocked; FP2, begun on 7 August 1998, reached 12 km of separation, met a thruster pulse injection failure during final approach, and docked after recovery on 27 August 1998; and FP3, on 26 and 27 October 1999, included an R-bar approach initiated 9 km from the target and remote-piloted sequences from 2 m out to hold points at 6 m and 12 m [2]. Every approach was to a cooperative target carrying LED markers and a corner cube reflector, and only three docking events exist.

Free-floating control in flight. Generalized Jacobian resolved motion-rate control maps joint rates to inertial-frame hand motion including the base reaction, and was flown [1]. In flight, a 200 mm straight-line inertial path tracked at 10 mm/s held hand orientation at -22.8 degrees in the inertial frame while the base pitched 0.6 degrees under the reaction moment [1]. Reaction null-space and zero-reaction motion, which finds arm motions that produce zero net reaction on the base, was validated in flight [5], [14]. The velocity-level redundancy resolution that preceded it is analytical and contains no experiment or simulation [12]. With a six degree-of-freedom non-redundant arm the null space available is n minus 3, so three of the six joint freedoms remain for reactionless motion, and attitude variation during the final approach of a capture was held to 0.5 degrees peak to peak [1]. The same formulation applies to a manipulator on a flexible base, where the target is suppression of structural vibration rather than rigid-body attitude change [9]; its original formalization rests on one planar simulation and no hardware [13].

Manipulation experiments. Four categories were flown: ground teleoperation under large time delay, orbital replacement unit exchange, truss deployment and assembly, and antenna assembly [4]. Antenna reflector assembly was performed by teleoperation with vision and task-level sequencing [7]. The hand was evaluated against the ORU exchange and truss handling tasks [6]. Hand-camera images of task board markers gave about 1 mm positioning, and the connector’s own pinhole pattern improved that to about 0.25 mm [6]. Capture of the free-flying target started with the hand about 400 mm from the grapple fixture against a target drifting at 10 mm/s, with the controller switching mode when hand position error fell below 100 mm [1]. Mechanical compliance was added to the antenna assembly mechanism because software compliance cannot follow motions faster than the onboard control cycle, and arm speed was held to 2 mm/s [7]. The mechanism corrected ten deliberately offset cases of plus or minus 5 mm and plus or minus 1 degree with a force never above 8 N, and completed 40 disassembly and reassembly cycles over a year and a half with no mechanical degradation [7].

Teleoperation under delay. Motion commands were uploaded at 4 Hz in an isochronous stream [1]. DLR’s GETEX experiment used task-directed programming with a predictive three-dimensional graphics display, inherited from ROTEX, in which the operator commands a simulated arm on the ground and the flight arm follows [8], [22]. The Kyoto and NASDA experiment instead closed a direct bilateral coupling with kinesthetic force feedback across the 6 to 7 second delay [3], the harder case because the delay sits inside the force loop. A slope-tracing task and an 18 mm peg-in-hole insertion with 0.4 mm clearance were performed without visual feedback, at a hand speed limited to 2.0 mm/s [3].

Later flights

DARPA’s Orbital Express (launched 8 March 2007, ended July 2007) flew the ASTRO servicer and NextSat client through mated and unmated rendezvous and capture, hydrazine transfer, and battery and computer ORU exchange by robotic arm; the program lead reports 100 percent of mission criteria met, a contractor self-report of one flight with one cooperative client [19]. The flight capture system had a 3 cm/s design contact velocity, 6 in axial capture distance, plus or minus 2 in lateral and plus or minus 5 degrees angular misalignment, and the vendor sheet gives no qualification data [20].

DLR’s ROKVISS two-joint torque-controlled arm has operated on the ISS Russian Service Module since February 2005, with telepresence and automatic-mode experiments and in-flight friction identification; it has two joints and no free-flying base [21]. KONTUR-2 closed a bilateral loop between a joystick on the ISS and robots on the ground with a measured round trip delay of 20 to 30 ms over the S-band link, across 23 space sessions from August to December 2015 [23]. Both are one to two orders of magnitude below the ETS-VII delay, so neither tests the case ETS-VII tested. Free-flyers inside the ISS have a different role: Astrobee has logged more than 165 real-time operations and more than 90 Guest Science test sessions, but it carries no arm in the ETS-VII sense [33].

What is being tried now

Dynamic-coupling and selective decoupling control. Work on free-floating manipulators continues to use ETS-VII as the parameter set. A multi-axis selective decoupling framework nullifies base momentum transfer only along mission-critical axes, leaving more joint freedom than full reactionless control; it is simulation only, with three point-to-point reaches and a rigid model [26]. A coupling-informed trajectory optimization for a planar 3-link arm uses the singular values of the coupling matrix and a sliding-mode tracker, in two simulated cases [25]. Both are simulation-stage proposals; neither has flown or been tested on hardware.

Learning-based and hybrid control. A data-assisted port-Hamiltonian controller for a planar 2-DOF free-floating arm reduced tracking error by 77.5 percent against one baseline in a single simulated run [28]. A trajectory-optimization and reinforcement-learning hybrid controls a free-flying four-arm 27-DoF servicing robot, entirely in a rigid-body simulator [29]. A real-time hardware-in-the-loop simulation platform validates a robust joint-space controller for a free-floating dual-arm servicer, with the servicer properties taken from a design concept, not flown hardware [32]. All three are simulation or simulation-in-the-loop.

Using the arm for attitude. Propellant-free slew and detumble maneuvers driven by the manipulator’s own joints, with joint-limit and collision constraints, are planned by trajectory optimization with rigid-body dynamics only [27]. It extends the ETS-VII observation that arm motion moves the base into a design tool. It is a simulation result.

Capture of non-cooperative targets. A coordination controller for a free-flying manipulator and a minimum-time detumbling trajectory were demonstrated for interception on a fixed-base laboratory arm with a mockup moved by a second arm; the detumble is simulation only [30]. Capture of a spinning object by repeated deliberate impacts from a dual-arm robot, which does not need the target mass or inertia, was demonstrated on a planar air-bearing floor with a cylindrical target, external motion capture and a three-degree-of-freedom arm on each platform [31]. A combined impedance and PD controller for a two-arm free-floating manipulator capturing a tumbling target has been studied in a multibody co-simulation only [34]. Grasping a spinning target with a free-flying robot has been addressed by an offline database of optimal trajectories queried over a four-dimensional description of the target motion, with simulation results only [35]. After capture, an adaptive reactionless control with online momentum-based inertia identification handles the changed dynamics of an unknown tumbling target; the published result is simulation [36]. A 2024 survey of visual servoing for on-orbit servicing finds most validation is in simulation or software-in-the-loop, and reports no numeric performance figures [17].

Ground emulation. A 2024 survey lists the facilities used for such work: ESA ESTEC’s ORBIT flat floor, DLR’s EPOS 2.0 for non-cooperative tumbling targets, NASA’s 21.3 by 29.9 m air-bearing floor at Johnson and 13.4 by 26.2 m floor at Marshall, the Langley Rendezvous Docking Simulator, and a two-manipulator hardware-in-the-loop testbed at York University [16]. DLR’s OOS-SIM carries a simulated servicer and client on two KUKA KR120 robots driven by free-floating dynamics at 4 ms, with a common workspace of about 23.28 m3 [37], and the EPOS 2.0 simulator uses two six-axis robots, one on a 25 m linear slide, to test rendezvous sensors and docking tools over the final 25 m to 0 m [38]. Air-bearing floors emulate two dimensions, so out-of-plane torques are not reproduced [31].

Programs and catalogs. NASA’s 2025 ISAM state of play catalogs the Core Module Manipulator and Experimental Module Manipulator, with brief entries that cite no sources for their figures [18]. Flexible long-reach arms raise a related problem; input-shaping filters for such arms have been compared in simulation only, for a terrestrial tank-waste manipulator and not a spacecraft [11].

The numbers

Quantity	Value	Source
Link delay, ETS-VII via TDRS	6 to 7 s	[3]
Link delay, KONTUR-2 via ISS S-band	20 to 30 ms	[23]
Docking lateral accuracy, measured vs required	7 mm vs 25 mm	[2]
Docking attitude accuracy, measured vs required	0.2 degrees vs 1.9 degrees	[2]
Docking approach velocity	1 cm/s	[2]
Orbital Express capture design contact velocity	3 cm/s	[20]
Base pitch under reaction, 200 mm path at 10 mm/s	0.6 degrees	[1]
Attitude variation, capture final approach	0.5 degrees peak to peak	[1]
Antenna assembly force, maximum	8 N	[7]
Hand positioning, marker / connector pattern	about 1 mm / about 0.25 mm	[6]
Bilateral teleoperation hand speed limit	2.0 mm/s	[3]

What is not established

The bilateral teleoperation result rests on one to three runs per condition with three operators, and its authors claim only that direct bilateral control at 6 to 7 s is possible, not that it is practical; operation was slow and limited to planar or low-degree-of-freedom tasks with an off-the-shelf joystick [3]. The comparison with the earlier NASDA compliance-mode work used different target forces, 5 N against 25 N, so it is not like for like [3]. The antenna assembly force is reported only as an 8 N maximum, with no distribution across the 40 cycles and no split between arm and mechanism [7]. The flight experiments of the arm control methods are single trials with most values read from plots, and a fair reactionless against point-to-point comparison over the same path was not possible because joint limits blocked a continuous path [14]. The docking record has three events, all with a cooperative target [2], and Orbital Express is likewise one flight with one cooperative client [19]. The public mission page gives launch and orbit parameters without experiment data [15]. Capture of a tumbling or uncooperative target, impedance control of a free-flyer and learning-based control have no flight record from ETS-VII; the published support is simulation, a fixed-base laboratory arm or a planar ground testbed [29], [30], [31].