Dextre

NASA. Public domain (NASA / US government work).

Overview

A station the size of the ISS accumulates failed boxes, worn cables and jammed hinges faster than a spacewalk schedule can absorb them, and every hour of extravehicular activity carries suit consumables, crew fatigue and depressurization risk that a robot does not. Dextre, formally the Special Purpose Dexterous Manipulator, was built to take that maintenance load off the crew: inserting and extracting orbital replacement units, actuating hinge-like mechanisms on the truss, and manipulating and inspecting payloads that would otherwise need a suited astronaut at the work site [1].

The requirement that shaped the design was commonality, not novelty. Dextre mounts on Canadarm2 or on a station grapple fixture and shares that arm’s Power Data Grapple Fixture interface and Latching End Effector design, so it can be carried by Canadarm2 or anchor itself directly to the station [3]. The Space Station Remote Manipulator System whose joints and end effectors Dextre’s interfaces were built to match was itself still a preliminary design when its mechanism specification was written, with detail design in progress and force-moment sensing still under investigation [9]; the convergence on shared interfaces across both arms was a program decision to avoid a second qualification campaign, not a discovery made through flight experience.

Specifications

Parameter	Value	Source
Arms	2, 7 joints each	[1], [2]
Arm reach	3.3 m straight-arm	[1]
Payload capacity per arm	600 kg	[2], [1]
Body roll joint	1 DOF at the waist	[2]
Motion resolution requirement	2 mm	[2]
Joints total, including the flight spare arm	21	[3]

Mission profile

Parameter	Value	Source
Host	International Space Station, on Canadarm2 or a station grapple fixture	[2]
Launch configuration	9 separate components
On-orbit assembly	about 6 h of EVA
Station orbital replacement units designated for robotic servicing	about 250	[2]
Of those, compatible with Dextre interfaces	about 200

Rows with no marker are from [3].

Each arm has three joints at the shoulder, one pitch at the elbow and three at the wrist [2].

End effector and tools

Each arm terminates in an ORU/Tool Changeout Mechanism carrying a parallel-jaw gripper compatible with H fixtures and micro fixtures, a retractable 7/16 inch socket drive, an extendable umbilical passing power, data and video to the grasped unit, a two-stop zoom camera and two lights [2]. A body-mounted tool holder carries four tools that the arms exchange without crew involvement, and two outriggers carry camera and light pan-tilt assemblies.

Payload interfaces are typed. The standard interface is a micro fixture, whose normal operating loads of 100 N and 85 N m bound the peak transient contact forces and moments allowed during a grasp; units with a micro-conical interface, such as the 6B avionics boxes, are handled instead through a separate Robot Micro-Conical Tool, whose offset makes the force-moment accommodation problem harder [1]. Heavier payloads are grasped directly at a grasp fixture. A dedicated H fixture on the station stabilizes the body during two-arm work [2].

Only one arm moves at a time, a power limit rather than a control limit [1]. The other holds a fixed grasp to provide a stable reference, since two arms sharing a compliant base would couple their position errors, and the 2 mm motion resolution requirement [2] does not survive that coupling.

Force-moment accommodation

Each arm carries a six-axis force-moment sensor on the OTCM [1]. Force-Moment Accommodation closes an outer loop on that sensor so the arm complies with contact rather than driving through it. Commanded steady-state limits are 4 N of lateral force and 3.1 N m of lateral moment, excluding transients, within a user-selectable range of 0 to 111 N and 0 to 54 N m [1].

Parameter	Value
Teleoperated positioning accuracy	6 mm, 2.0 deg
Automatic incremental adjustment	2 mm, 0.1 deg
ORU capture envelope	5 cm lateral, 8 deg about any axis
Micro fixture operating limits	100 N, 85 N m
Guide design load	50 N, 50 N m
Transient force limit during bolt following	12 N peak
Insertion force and moment resolution	4 N, 2.37 N m
Gripper jaw closure	15 to 20 s
Operator guideline on peak transients	5 s maximum duration

Source: [1].

The case for the control law is measurable. Three operators inserted a DDCU orbital replacement unit on the SPDM ground testbed with and without Force-Moment Accommodation [1]:

Operator	Without FMA	With FMA
1	1.5 mm/s, 5 min, peak 200 N and 12 N m	5.5 mm/s, 2.0 min, peak 18 N and 25 N m
2	0.5 mm/s, 17.5 min, peak 190 N and 45 N m	5.5 mm/s, 2.0 min, peak 19 N and 24 N m
3	1.5 mm/s, over 15 min, peak 220 N and 37 N m	4.2 mm/s, 2.5 min, peak 16 N and 20 N m

Source: [1].

Peak contact force falls by roughly an order of magnitude and insertion time by a factor of three to eight. Since orbital replacement units cannot be repaired on orbit, the peak force column is the binding number, not the time column, and the requirements verification program was built around demonstrating those interface loads rather than around task speed [3]. Ground testing is run inside a gravity envelope of plus or minus 3 degrees of attitude, within which gravity compensation holds residual force below 3 N and residual moment below 1 N m [1]; the testbed travels 5 ft vertically, 3 ft along the insertion axis and 1 ft laterally.

Thermal

Dextre and its payloads are maintained by heaters sized against eclipse. The Robotic Refueling Mission payload it hosts illustrates the allocations: two Shuttle heater feeds totalling 230 W at maximum voltage, about 216 W at nominal, each circuit controlled by two Elmwood mechanical thermostats in series switching between +25 and +30 C [4]. On ELC-4 the station provides two feeds totalling 290 W at maximum voltage, about 245 W nominal. Tool heaters are separately powered through the OTCM: 5 W per tool camera with two cameras per tool, and 10 W per tool electronics box, with thermostat set points of -25/-16 C for the cameras and -32/-26 C for the electronics [4].

Robotic Refueling Mission

Dextre hosted the Robotic Refueling Mission series, which operated on representative satellite fuel valves including wire-locked caps, cutting safety wire, removing caps and transferring fluid [5]. RRM3 extended this to cryogenics.

RRM3 parameter	Value	Source
Source and receiver dewar volume	50 L and 10 L	[6], [19]
Methane load at launch	about 19 kg	[6]
Zero boil-off duration, ground plus orbit	6 months
Demonstrated on-orbit storage	4 months
Receiver dewar temperature	below 82 K
Methane triple point	90.7 K
No-vent transfers demonstrated on the ground	5
Autogenous pressurization cycle	20 min
Ground supply dewar	450 liter DOT-rated Chart Microfueler	[7]
Methane purity requested / delivered	99.9 / 98.2 percent	[7]
Launch	5 December 2018	[7]

Rows with no marker are from [6].

Ground servicing with liquid methane imposed a 150 ft dynamic hazard control zone during testing, a 50 ft continuous control zone, a vent stack 35 ft from the work building and an 18 ft standoff for the command trailer [7]. Machine vision calibration for RRM3 held below 2 mm of error [5], and telerobotic operation required a minimum of 10 frames per second of video, with 24 preferred. Operations ran 16 to 20 hours per day with a three-person flight controller team and about twelve supporting staff at Goddard.

A related payload on the same servicing project, the Dextre Pointing Package, would have mounted rendezvous and docking sensors (a triangulation and time-of-flight lidar, a star tracker and a 6-DOF inertial measurement unit) on the arm to give closed-loop tracking of visiting vehicles on-orbit exposure that a laboratory testbed cannot reproduce, since a realistic rendezvous envelope spans kilometers of range and nearly the full sphere of attitude, tractable in a lab only below about 25 m [10]. The package did not fly.

The Canadian Space Agency operates Dextre jointly with NASA and is the organization responsible for its public technical information [12].

Ground operations

Round-trip communication delay to the station is 5 to 6 seconds or more [8], so continuous force-reflecting teleoperation is not available and the Force-Moment Accommodation loop has to close on orbit rather than on the ground. Procedures are developed against a six-degree-of-freedom industrial robot at Goddard and an MDA ground trainer carrying SPDM-representative hardware [5]. A parallel line of work proposed running station arm operations from the ground entirely, driven by a task-level programming and control framework rather than direct teleoperation, on the reasoning that crew time is the scarcer resource when an arm moves slowly over long displacements [8]; that architecture was demonstrated only against a simulator, and nothing in the corpus shows it flown against the Mobile Servicing System.

Robotic cargo operations are long by EVA standards but cost no crew time. Dextre’s role inside the broader on-orbit servicing landscape, alongside other short-reach station manipulators, is one entry in NASA’s periodic survey of servicing, assembly and manufacturing capability, whose 2021 edition set the capability taxonomy that later editions still use [11].

Place in the Mobile Servicing System

Canadarm2 supplies the reach. The agency table gives it 17 m, 1497 kg, seven joints with 540 deg of rotation each, and tip speeds of 37 cm/s unloaded and 2 cm/s loaded under ground control, against 15 m and 410 kg for the Shuttle arm [15]. Its seventh joint makes it kinematically redundant and symmetric about the elbow, so either end can be the base, which lets it walk between power-data grapple fixtures,. Closed-form inverse kinematics for the arm split the arm into three six-joint subchains with at most eight configurations each, solved by quadratics [14]. A 1987 trade study for the Mobile Transporter that carries the arm along the truss fixed the requirements to traverse every truss face and handle payloads over 100,000 kg, at a time when the arm design was still open [30]. The Spar simulation facility built for the Canadian Space Agency modeled both the SSRMS and the SPDM with flexible multibody contact dynamics, but reports no validation error [13]. Operator interfaces were compared on a Space Station mockup in 1994 with ten astronauts, a controls and displays test that measured nothing about arm performance [29].

Heritage and drive components

Dextre was verified largely as an assembly of previously qualified parts. The Shuttle arm end effector it descends from closes a three-cable snare by turning a ring through 80 deg and tolerates 4 in of lateral and axial misalignment, with the figures given as design requirements [28]. A 1987 survey of electromechanical drives for space manipulators compared ten transmission families on backlash, friction and stiffness, from untraceable prior tests [31]. The European Robotic Arm, the other station arm built to the same generation of interface thinking, uses a redundant motor driving a 454 to 1 epicyclic stage with torque ripple cut from roughly 15 percent to under 0.3 percent, qualified at unit level before the arm flew [27].

Other manipulators and servicing approaches

Dextre is one of several flown or planned station manipulators, and the sources group them by what they were built to handle [21], [22].

• Large station arms. The European Robotic Arm serves the Russian segment because Canadarm2 and the Japanese arm cannot reach its base point and payload interfaces [26]. The Chinese station carries a core module arm and an experiment module arm; the design paper gives the core arm 10.5 m, 738 kg and 25,000 kg payload [24], while a state-run article gives 10.2 m [25], and NASA’s catalog carries both arms [23].
• Humanoid and free-flying robots. Robonaut 2 flew with 42 degrees of freedom, six checkout operations and first crew teleoperation in April 2013, with no task success rates reported [19]. A SPHERES docking and servicing payload set was tested only on a flat floor and in a reduced gravity aircraft [35]. Neither replaces external maintenance.
• Crew and ground teleoperation. Bilateral force-feedback control from the station to ground robots measured 20 to 30 ms round trip over S-band across 23 sessions [20], and three astronauts drove a rover over 11 hours from the station [34]. Dextre’s 5 to 6 s ground delay rules this approach out for contact operations.
• Next-generation arm. Canadarm3 for the Gateway is a proposed 8.5 m, 715 kg (estimated) design that has not been built [15], [33].
• Propellant transfer. A heated-wick pressurization concept for RRM3 was developed with liquid nitrogen at one g [18], and a hydrazine transfer subsystem for a servicing spacecraft was characterized with surrogate fluids [32].

RRM3 held methane for 165 days without venting, but its five ground transfers moved 9.5, 1, 8.5, 0.8 and over 10 liters, the second and fourth failing and the fifth ending in a liquid lock that ruptured a burst disk [17]. No cryogen was transferred on orbit.

What is not established

The Force-Moment Accommodation comparison is a single run per operator per condition, run without repetition because operators did not want to repeat the unassisted case, so the without-FMA numbers are as much a record of operator frustration as of a controlled baseline [1]. The SPDM ground testbed’s claimed match to a zero-gravity simulation rests on one comparison plot described only qualitatively, with no residual error reported [1]. The interface commonality argument between Dextre and Canadarm2, that clearing shared items by environment comparison alone is sufficient, is stated as the verification plan rather than shown to have worked, since the paper describing it was written while verification was still underway [3]. The proposed ground-control architecture for the Mobile Servicing System has no flight demonstration and no reported task times, delay compensation error or operator study [8]. RRM3’s central objective, cryogenic transfer in microgravity, was itself never exercised: fluid management at the transfer line inlet, no-vent transfer and freezing behavior of methane in zero g remain unanswered, and most of the ground campaign substituted argon and nitrogen for methane because of facility limits [6]. The Dextre Pointing Package’s pointing and sensor performance figures are simulation only, since the payload never flew [10]. The Canadarm3 figures are design targets, and the agency tables for Canadarm2 give no payload mass for the quoted speeds [15]. The agency pages carry no independently verified specification [16], and the two Chinese arm length figures are unreconciled [24], [25].