Rollin' Justin

DLR. CC BY 3.0.

Overview

Rollin’ Justin is a wheeled humanoid built by the DLR Institute of Robotics and Mechatronics. It is the ground element of two ISS telerobotics experiments that between them tested the two ends of the command spectrum: KONTUR-2, in which cosmonauts flew a force-feedback joystick and drove ground robots as avatars, and METERON SUPVIS Justin, in which astronauts supervised the robot as a co-worker through a tablet. The two campaigns together covered teleoperation of robots on Earth from orbit across four years, 2015 to 2018 [2].

The upper body was an existing two-arm humanoid; the 2009 addition of a mobile platform made it Rollin’ Justin [1]. The upper body carries 43 degrees of freedom and the platform adds 10 to 12.

Specifications

Parameter	Value
Total mass	199 kg
Upper body mass	about 50 kg
Platform mass	about 150 kg
Upper body degrees of freedom	43
Arm span	3000 mm
Lift capacity	20 kg
Torso travel	600 mm forward, 300 mm back
Battery	48 V nominal Li-polymer, 40 Ah
Average power draw	about 600 W
Endurance	about 3 h
Top speed	1.5 m/s
Upper body control rate	1 kHz

Source: [1].

The DLR program page states figures that differ from the platform design paper: 51 degrees of freedom for the whole system, distributed as 8 in the platform, 7 in each arm, 12 in each hand, 2 in the neck and 3 in the torso; a standing height of 1.91 m; a working range from floor level to 2.7 m; a maximum speed of 2 m/s; and an operating time exceeding 60 minutes [11]. Nominal load capacity of 20 kg agrees between the two [1], [11].

Mission profile

Parameter	Value	Source
Role	ground element of ISS telerobotics experiments, commanded from orbit
KONTUR-2	cosmonauts driving ground robots through a force-feedback joystick
METERON SUPVIS Justin	astronauts supervising the robot through a tablet
Campaign span	2015 to 2018
First public presentation	2008	[11]
Current role	research platform in service robotics	[11]

Rows with no marker are from [2].

Mobility

The platform is a four-wheel omnidirectional base with a variable footprint. Each wheel unit carries an independent brushless DC hub motor rated at 30 N m driving a steel rim with a 5 mm solid rubber tire of 216 mm diameter, giving 1.5 m/s (5.4 km/h) [1]. Steering is a Harmonic Drive FHAC-mini brushless DC motor and 100:1 harmonic gearbox reaching 360 deg/s and 28 N m, run by an Elmo Whistle digital servo-drive with a Renishaw RE22S absolute encoder reading the gearbox output. Rollin’ Justin passes steps up to 40 mm and climbs ramps to 17 degrees [1].

The footprint changes because manipulation and locomotion want different geometry. Extended, the wheelbase is 985 by 815 mm, which with a center of gravity about 260 mm above the ground is enough to prevent tipping during an emergency stop at full speed with the upper body bent forward; retracted it is 685 by 515 mm, which lets the platform go anywhere a wheelchair can go [1]. Overall dimensions run from 1220 by 1052 mm to 812 by 644 mm, with height varying between 658 and 728 mm on the suspension and 693 mm in the neutral state. The extension mechanism is passive: a parallel linkage moves each wheel horizontally without vertical displacement, actuated by the wheel motors themselves rather than by dedicated actuators, and held by a toothed-rack lock driven by a two-position servo. Each leg carries a spring damper allowing 35 mm of deflection about equilibrium so that four wheels stay in contact on uneven ground and doorsteps are crossed gently [1]. For SUPVIS Justin this platform is what let the upper body navigate the Solex ground environment and carry tools between work sites [2].

Torques on the last torso axis range from 248 N m static to 1800 N m under load, which is what set the stability requirement on the base [1].

Power and energy

The battery is a 48 V nominal lithium-polymer pack of 13 cells in series at 3.7 V each, 40 Ah capacity, with a maximum discharge current of 200 A and a maximum charge current of 80 A [1]. Terminal voltage swings between 40 and 54 V with state of charge. Against an average whole-robot draw of about 600 W the pack runs for about three hours, and recharges in one hour at maximum charge current [1]. A balancing and safety board monitors and corrects voltage, current and temperature per cell, and the pack sits behind a stainless protective cover because of the energy density.

Six separately fused supply lines distribute 48 V, three rails of 24 V, 12 V and 5 V, with current measured per line and displayed on a status panel; the supply circuit handles 3.7 kW continuous [1]. Because the platform runs untethered a wireless emergency stop is required: a radio and infrared receiver drives a current loop carrying an OK signal to the motor controllers and supplies, with the transmitter sending on both 433 MHz and 868 MHz and over infrared so that a single disturbed path does not defeat it, and a receiver timeout of 200 ms. Onboard battery endurance is what set the session length available for ground work during the ISS campaigns [2]. The design budget allocated 1 kW to actuation plus at least two computers, with 15 to 50 W per hand, 50 to 150 W per arm and 100 to 250 W for torso and head [1].

Thermal

No thermal control system is published for Rollin’ Justin itself; several fans cool the platform interior [1]. The flight hardware of KONTUR-2, the joystick, is the part with a documented thermal design. It is passively cooled to two constraints: the housing must never exceed 40 C, because the cosmonaut touches it, and the electronics must not overheat [3]. Heat generating parts are therefore placed at maximum distance from touchable outer surfaces and thermally tied only to the bottom plate, with space-qualified gap pads and graphite foils conducting heat into the mechanical structure and suppressing hot spots, optimized against finite-element models. The model was built in ANSYS 14.0 against an ISS ambient of 28 C or below, complete insulation between joystick and ISS structure, heat flows carried with 10 percent margin, an electroplated chromium housing at emissivity 0.1 and a black anodized aluminum adapter plate at 0.82, and convection from the 0.05 m/s ISS supply airflow near the unit [9]. Dissipation by state is about 1.9 W from the DC/DC converter and about 5.0 W from the microcontroller module in every state, with the motor control modules adding about 3.5 W in idle and a load-dependent figure in operation.

Because real motor load depends on the task and on how the cosmonaut handles the joystick, a software temperature control system reads nine calibrated sensors at 0.1 Hz, intervenes automatically, and gates the operator’s state changes on a temperature criterion, dropping to standby when the unit is too hot [3]. An OpHalf state permanently halves the commanded handle force to extend running time before the housing limit is reached, against a required operating time of 30 minutes.

Compute and avionics

Four MiniITX standard PCs with dual-core processors sit in the platform for control, communication and peripheral I/O, linked by Gigabit Ethernet through an eight-port switch, with a second eight-port switch for the video and PMD cameras and a Wireless-LAN bridge as the uplink to external control [1]. About 50 liters of chassis volume holds the battery, power supply, computers and infrastructure.

Bus choice follows the component rather than the other way round, because the platform was built from off-the-shelf parts selected on size and function [1]. The hub motor wheels run CAN 2.0B at a 16 ms command rate; the Whistle steering servo drives run CANopen at 4 ms; absolute encoders for steering angle and leg extension report over SSI; the arms, torso and head run the SERCOS real-time bus. The upper body control loop runs at 1 kHz [1].

The KONTUR-2 joystick avionics are a separate design point, built to ISS rules. Safety-critical parts use highly reliable components while computing and motor electronics are military, industrial or automotive grade [3]. The microcontroller module is a commercial part with a processor and an Ethernet switch, sized for the force feedback controller, the operational state machine, error handling, I/O, temperature supervision and communication with the motor modules and the Portable Control Computer. Motor control modules speak EtherCAT to it. Brushless DC motors were chosen both for ripple-free constant force and to avoid carbon brush wear polluting cabin air. Input power is 23 to 29 V so the nominal 28 V Russian segment supply connects directly, behind a power switch, EMC filter and two-stage overcurrent protection whose current limiter also soft-starts to bound inrush [3].

Software runs VxWorks 6.9 with two boot paths against radiation-corrupted memory: bootloader and kernel come from the first SD card partition by default with an alternative pair on flash, selected on CRC32 checksums held in EEPROM [3]. The application is startable from SD card, FTP folder or flash in that order, verified by Simple File Verification and revalidated every minute [3], [8]. Above three kernel-space drivers sits the HIROSCO component framework, which manages access over ECSS PUS and exposes a TCP/IP server for monitoring and control.

Autonomy

SUPVIS Justin treats the robot as a co-worker: the robot’s own intelligence plans and executes low-level tasks while the astronaut supervises, issuing high-level commands and checking status and progress [2]. Whole-body control gives compliant interaction with the environment, and reasoning and planning are implemented as Action Templates. For each known object a template holds the properties and preconditions of every action available on that object together with the geometric procedure for carrying it out, so joining templates lets the robot plan and execute a complex task from a single astronaut command.

Because models of known objects can be loaded in advance, the robot’s computational load is eased and the task set is bounded, which is what makes the approach suited to a planetary surface habitat where the assets were sent from Earth and are therefore known [2]. The architecture is built so that communication delay and jitter arising from distance and disturbance do not gate the task, because the astronaut commands at task level rather than at joint level [4].

The command list a template set generates can be large. A Mission Control Center on Earth prunes it context-specifically with symbolic and geometric filters, steering the astronaut toward task completion without taking the command away [2].

KONTUR-2 is the opposite arrangement. The human closes the loop directly, and the robot carries no task-level autonomy at all; the design problem is stability under delay rather than planning. Time delay, jitter and data loss are handled by the Time Domain Passivity Control Approach [6]. The DLR side used a four-channel bilateral architecture with both joystick and ROKVISS impedance controlled, position and velocity sent one way and computed and measured forces the other; RTC used a dual-channel architecture with its own controllers for Surikat.

Communications

The two experiments used different links [2].

	KONTUR-2	SUPVIS Justin
Format	point-to-point	via TDRS
Band	S-band	Ku-band
Uplink	256 kbit/s	10 Mbit/s maximum
Downlink	4 Mbit/s	10 Mbit/s maximum
Round trip	20 to 30 ms	about 820 ms
Availability	10 minutes per 90 minute orbit during direct fly-over	continuous link theoretically possible

Source: [2].

KONTUR-2 reused the S-band antenna designed for the ROKVISS experiment [6]. The SUPVIS Justin path runs ISS to TDRS to a US ground station and onward to Europe over the internet, and it is the gateway count and the distances that produce the larger delay; the compensation is that sessions can run for hours rather than for a fly-over [2]. An HD camera on the ISS let the ground team watch the astronaut, with voice on the ISS voice loop.

Ground training for KONTUR-2 ran over the internet between Star City and Oberpfaffenhofen at a mean delay of 65 ms, higher than the space link itself [6]. RTC sessions were connected to the ISS by paths other than S-band alone and saw increased delays, high packet loss and higher jitter, and the system still worked.

Payload and instruments

Sensing on the platform is arranged for autonomy rather than for science. Two color video cameras, one facing front and one facing left, provide visual odometry from optical flow, giving a redundant odometry solution with reduced error [1]. Four photonic mixer device cameras look front, left, right and back, returning a depth value per pixel for a 3D view of the surroundings that resolves walls, doors, obstacles and moving objects such as human legs. Every wheel carries forward and backward bumper switches as a last resort, since the wheels are the outermost parts, while the area between the legs is covered by the PMD cameras. A force sensor in each leg reads 0 to 1000 N, so weight on each wheel gives the center of gravity, and an inertial unit sits in the head [1].

For SUPVIS Justin the robot carried an assortment of electronic and mechanical tools [2].

The KONTUR-2 flight hardware is a two degree of freedom force feedback joystick, built around a cardan joint with a brushless DC motor and cable capstan reducer per axis [3]. It produces up to 15 N with a range of plus or minus 20 degrees on both axes, and the handle is inclined 15 degrees forward along its long axis to sit inside the wrist’s radial and ulnar abduction limits. A deadman button at the handle means the motors only apply force while the handle is gripped, disabled in software only for calibration and functional test; seven LEDs on the top face report operational state and error conditions. Driver-level safety adds a Cartesian force limit, temperature supervision, a power-on handle position calibration, a position controller for autonomous handle motion and an alive check [3], [8].

Crew-safety engineering is visible throughout. All power cables are double insulated in materials meeting ISS offgassing and flammability requirements, connectors have metallic housings, the unit is fully enclosed in a metal housing bonded to ISS structural ground with all internal parts galvanically isolated from case ground so that an accidental short to case cannot produce harmful current, and every board is conformally coated [3]. A protection cap guards the handle in transport and non-operation, qualified by impact tests at 556 N over 0.3 to 1.5 s, and in operation it mounts to the rear of the housing as an armrest. The transport bag is foam under flame-retardant Nomex, covering shock and random and sinusoidal vibration from 20 to 2000 Hz on all three axes, and measured to cut peak accelerations by a factor of about 24 [3].

Modes of operation

The joystick control loop samples at 1000 Hz, standard for haptics and long enough to carry the real-time controllers; the intrinsic delay from a torque command to its effect on the handle is 1 ms, and with the encoder resolution this gives a maximum stable stiffness of 1.57 N m/rad [3].

KONTUR-2 ran ten-minute sessions through 2015 and 2016 [2]. Five cosmonauts took part; three ran on-board studies against a local simulation and two commanded ground robots. Oleg Kononenko and Sergey Volkov performed the ROKVISS tasks [6]. Four ground robots were flown against: ROKVISS, a two degree of freedom torque-sensing arm used for peg-in-slot and contact work; Surikat, an RTC kinematically redundant manipulator with a stylus and light targets and no torque sensors, so force feedback was model-based; Yula, a small RTC mobile robot needing position-speed rather than position control; and Space Justin. Tasks covered tracing physical geometric profiles, following light patterns on an LED grid, grasping and manipulating an inflated beach ball, and shaking hands with a person on the ground [2].

SUPVIS Justin ran three ISS-to-ground sessions of four hours between 2017 and 2018 with five astronauts, each session carrying more complex tasks than the last [2]. The first session, on 25 August 2017, validated system usability with data readout, system reboot, survey and navigation and involved three astronauts [5], [2]. The second, on 2 March 2018, added dexterous manipulation including solar panel readjustment and dust wiping, with NASA astronaut Scott Tingle commanding the robot to clean a solar panel [5]. The third added component retrieval and mechanical assembly for a full end-to-end installation [2]. Six experiment protocols were designed for the five crew members and every protocol run in all three sessions was completed, some under tight time constraints from ISS crew scheduling.

Ground operations

The ground environment is Solex, the Simulated SOLar Farm EXperimental setup at DLR, built with the components expected in a future Martian habitat and extended through the experiment life cycle; the final configuration has solar panels on smart payload units, antenna receivers and a lander unit [2], [7].

The astronaut interface is a tablet. A high-resolution feed from the robot’s camera lets the operator see through the robot’s eyes, known objects are overlaid with their 3D models, and the Action Template commands are offered by point and click [2]. The interface was revised between the first and second sessions on the basis of crew performance and feedback [5]. Command options are presented on a dynamically updated interface that offers the currently feasible actions and returns execution feedback from the robot, across a catalog of survey, navigation and repair tasks [4].

Technologies developed

The two campaigns produced a comparison rather than a single result. KONTUR-2 showed that a multi-degree-of-freedom force reflection joystick can be paired with robots of different formats and complexities, delivering crisp force reflection thanks to the 20 to 30 ms delay and the controller’s delay handling [2]. The handshake was carried out with both the cosmonaut and the person on the ground reporting they felt the forces the other exerted, which was the first time a multi-DOF haptic input device was paired with a multi-DOF robot as an avatar to give a dexterous sense of physical contact between two people. It also demonstrated task-specific mapping of a two degree of freedom input device onto a seven degree of freedom arm.

The cost is operator load. Astronauts described the avatar modality as mentally and physically tiring, and it can only be run for 20 to 30 minutes before performance degrades, which was not a constraint for KONTUR-2 because each session gave only about ten minutes anyway [2].

SUPVIS Justin produced the opposite profile. The stated aim was to reduce the astronaut’s workload by delegating tasks to the robot, against the high mental and sometimes physical workload of direct teleoperation [4]. Astronauts consistently rated the supervised autonomy system easy to use and not a heavy mental workload, and all of them rated the concept capable of letting one crew member manage a team of multiple robots [2]. Crew with prior ground training completed the same task noticeably faster; one crew member with no prior knowledge was recruited on the spot and completed the protocol after interactive training from another crew member; speed and confidence rose visibly across the three sessions.

The conclusion drawn from the pair is a division by task rather than a winner. The avatar suits exploration, unknown environments and unexpected situations, where human intelligence and dexterity carry the command down to a low level. The co-worker with supervised autonomy suits large-scale tasks over known objects, where ease of use and low mental workload matter more than immediacy [2]. Both rest on the same finding: an operator in orbit above the surface has a low enough delay for either method, while keeping the crew out of the surface environment.

The METERON program continued past both. Analog-1 returned to haptic control of a surface rover from the ISS, taking the touch-based modality that KONTUR-2 established into a rover task [10].

References