FEDOR / Skybot F-850

Post of Russia (Pochta Rossii) issue; public domain in Russia per the Wikimedia Commons file page.

Overview

FEDOR, from Final Experimental Demonstration Object Research, is an anthropomorphic robot built by NPO Android Technics of Magnitogorsk. Modified for flight it received the call sign Skybot F-850 and flew to the ISS aboard the uncrewed Soyuz MS-14 as the first stage of the Roscosmos space experiment Ispytatel (Tester), which ran from 22 August to 7 September 2019 [1].

The robot itself predates the mission. It was developed under the project “Development of technology for creating a combined control system for robotic complexes”, code name Spasatel (Rescuer), for collaborative control using fine manipulator motor skills, movement through typical urban development, obstacle negotiation and vehicle driving [1]. The ISS experiment asked whether such a machine could be integrated into a crewed spacecraft at all, whether it would survive the flight environment, and what happens to teleoperation when the operator is himself weightless.

Specifications

Parameter	Value	Source
Height	184 cm
Shoulder width, ground robot	52 cm
Shoulder width, space version	48 cm
Mass, ground robot	160 kg
Mass, space version	105 kg
Degrees of freedom	48	[3]
Servo drives	48
Manipulator payload	10 kg each	[3]
Power plant rating, ground robot	13.5 kW
Spacecraft electrical interface, space version	up to 250 W average, 400 W briefly	[3]
Autonomous endurance on internal battery	1 h
Design life	up to five launches to orbit	[3]

Rows with no marker are from [2].

Mission profile

Parameter	Value	Source
Launch vehicle and spacecraft	uncrewed Soyuz MS-14
Experiment	Roscosmos space experiment Ispytatel, first stage
Duration	22 August to 7 September 2019, 16 days on orbit
Required operating time in flight	up to 5 h	[3]
Required operating time after landing	not less than 30 min	[3]

Rows with no marker are from [1].

Mobility

The flight configuration carries 48 degrees of mobility, distributed as 6 in each support module, 7 in each manipulator, 6 in each gripper, 7 in the body module and 3 in the head module [3]. Each manipulator lifts 10 kg. Geometric parameters match those of an adult human closely enough that the robot can be seated in a Cheget-type couch and held by the seat restraint system or by a purpose-built fixation system. The class of tasks the design targets is inspection, installation and servicing of equipment, and assisting cosmonauts during extravehicular activity.

The robot walks and drives on the ground. Published capabilities include opening a door, working with a drill and other tools, sitting in and driving a car and a quad bike autonomously, negotiating a slalom maze and climbing steps [2]. None of this was exercised on orbit: for the whole of the ISS phase the robot was rigidly fixed in the MIM-2 module and moved only its arms [1].

Later development iterations changed the kinematics substantially. A technological mock-up described in 2023 has 22 degrees of mobility excluding grippers, 8 per manipulator and 6 per gripper, on a structure of hull module, support modules, head module, communication system module, energy supply system, navigation system, stabilization and dynamic equilibrium system, sensor system and onboard computer [5]. Those figures describe the successor design, not the robot that flew.

Power and energy

The ground robot’s power plant is rated at 13.5 kW, with dual supply, through a cable or from a built-in accumulator [2]. For the space mission two built-in batteries from the Orlan spacesuit were used, and endurance in autonomous mode is one hour [2]. That rating is the robot’s own installed capacity and is not what the spacecraft delivers: the electrical interface between Soyuz MS-14 and the robot was budgeted at up to 250 W on average with brief maxima to 400 W [3].

Thermal

The robot is fitted with a heating system for important units, to protect against rapid cooling at low temperatures [2]. One of the stated objectives of the space experiment program was to study the heat release modes of the robotic system at its different stages [3]. No results from that study are published, and no radiator or coolant loop is described.

Compute and avionics

The system as flown comprises three elements: the anthropomorphic robot, a copying-type master device with a virtual reality helmet, and a controlling onboard laptop, an HP ZBook, running the control software [1].

Onboard instrumentation includes an inertial MEMS module and a system for recording the loads on the robot’s links and modules [1], [3]. Communication during the Soyuz phase ran through the spacecraft’s radio equipment, and a space communications headset was installed on the head module as part of the flight modification.

Actuation in the later design uses brushed DC motors chosen for small size at high specific performance, with planetary motor-reducers and belt drives in the manipulators [5]. The gripper actuating link groups use a Maxon Motor brushed motor driving a ball screw pair, with position read by an AS 5045 sensor and control by NPO Android Technics controllers. That drive reduces 4.4:1, feeds at up to 90.9 mm/s, and produces 64 N continuous and 176 N intermittent force [5].

Autonomy

FEDOR is a teleoperated machine with an autonomous mode, not an autonomous machine. The operating mode of interest for the experiment was copying control, in which the operator wears an exoskeleton master device and the robot reproduces his motion [1]. The rationale given is that copying control is most effective in situations of uncertainty, where the goal and the activity algorithm have to be adapted flexibly to changing external conditions, which is what a human is adapted to [1], [4].

The control technology set developed for the family covers fine motor control, reading and transmitting operator motion to form commands, force-moment feedback, operator immersion with augmented reality, gesture control, copying control, specialized instrument manipulation, filtering and holding modes, local navigation, dynamic gait balancing, autonomous control in a non-deterministic environment, leader-follower modes, and sensor-based control with feedback [6].

Force-moment sensors are fitted in the gripper modules and on the first links of the manipulator modules, to give the operator load information during copying control [5]. The control loop is framed as operator to robot to environment, with force-moment interaction closing it. The exoskeleton master device that closes this loop, called ZUCT, is a lever system with a structural scheme matching the human body; encoders on its joints register operator motion and generate the robot’s control signals, and active drives in the device return the robot’s own force sensor readings to the operator as proportional loads on his body [7]. Before the flight the ground robot completed 52 test operations under this control scheme, 26 of them worked through the exoskeleton across every manipulator degree of freedom, and the design team judged copying control the most rational mode available on that evidence [7].

Compared to other flown teleoperated humanoids

FEDOR is not the first humanoid built to work in space, and its design premise differs from the one NASA settled on for Robonaut. Robonaut was built to avoid a robot-specific worksite: rather than requiring its own approach corridors and interfaces the way the Space Station’s dedicated manipulator does, it fits the same corridors, uses a subset of the EVA tool set, and can respond to an EVA emergency far faster than the hours of prebreathe an astronaut needs [8]. That is a design argument about infrastructure, not the task-uncertainty argument FEDOR’s own program makes for copying control, where session length rather than task type is the ergonomic limit the ground program measured [4]. Robonaut 1 reached 47 degrees of freedom across two seven degree of freedom arms, two five-fingered hands and a three degree of freedom waist, sized to the torque needed to remove and install an orbital replacement unit, up to 20 lbf and 30 in lbf, and by 2001 had demonstrated a climbing traverse on a Space Station module mockup while mating a connector at each end [8], [9]. Robonaut 2, delivered to the ISS in 2011, moved to series elastic actuation with custom torsion springs at each upper-arm joint for fine torque sensing without giving up payload, and closes a 10 kHz torque loop in embedded motor drivers under a central kinematics and safety controller [10]. FEDOR’s own actuation is simpler: brushed DC motors with planetary reducers and belt drives, and a gripper drive using a ball screw pair read by a position sensor rather than joint-level torque sensing [5].

The teleoperation link is the sharper contrast. DLR’s KONTUR-2 and METERON SUPVIS Justin missions flew nine astronauts and cosmonauts commanding ground robots from the ISS between 2015 and 2018 [11], at each extreme of coupling: a 2-DOF force-feedback joystick making the ground robot an avatar of the crew member, and a tablet making it a supervised coworker. High-coupling telepresence over KONTUR-2’s point-to-point link, 20 to 30 ms round trip and about 10 minutes of contact per orbit, was judged tiring enough to limit useful sessions to 20 to 30 minutes [11]. FEDOR’s copying sessions, run through the Soyuz and ISS radio equipment rather than a dedicated telepresence link, averaged about one hour per operator in ground rehearsal [4], and the session structure driving crew time in flight came from that ground figure rather than from a link budget. None of the design lineage behind FEDOR traces to the JPL humanoid program that argued, in the mid-2000s, for humanoids as partners in lunar and planetary base construction and laid out a decades-long capability roadmap against Project Constellation [12]; nor to the broader orbital-manipulator and rover tradition a general space robotics survey places it against, in which free-floating dynamics and wheel-terrain mechanics rather than anthropomorphic form are the shared mathematics [13]. FEDOR is a Roscosmos line with its own precedent, the Rescuer program’s copying-control research, tested on a task set of building-relevant motor skills and vehicle driving rather than EVA support, and the KONTUR-2 comparison is where that difference of aim shows up most directly in the numbers [11].

Communications

Voice and data during the Soyuz MS-14 phase ran through the spacecraft’s radio equipment, and the voice channel with the built-in space communications headset was qualified on a flight Soyuz MS before launch [1], [3]. No band allocation, data rate or latency figure has been published for the teleoperation link.

Payload and instruments

The robot carried no science payload. Its instrumentation exists to record its own flight environment and to support control: an inertial MEMS module logging linear accelerations, angular rates and magnetic field induction on three mutually perpendicular axes in the global frame; cameras and microphones writing video and audio to non-volatile memory; and the link and module load recording system [1], [3].

On orbit it worked with the crew’s own hand tools rather than with dedicated robotic tools, which was the point: the operations tested were the ones cosmonauts perform when repairing and replacing equipment aboard the Russian segment [1].

Modes of operation

Two modes were exercised. During ascent aboard Soyuz MS-14 the robot ran autonomously, executing a fixed program: reporting the stages of ascent and the recorded overloads, and writing video, audio and inertial data to non-volatile memory [1]. On the station it ran under copying control from a cosmonaut. The environment the autonomous mode had to ride out was specified in advance as overloads of 5 g nominally and up to 16 g on atmospheric entry, impact loads on landing, cosmic radiation, acoustic loads, electromagnetic interference from spacecraft equipment, and the atmospheric factors of a pressurized compartment [3].

The ascent record is the most complete engineering result the experiment produced, because the robot instrumented its own launch. Five phases appear in the axial acceleration data, with Z along the vehicle longitudinal axis [1]:

Phase	Maximum Z acceleration
On the launch pad	1 g
Launch and first stage	3.89 g
Second stage	2.53 g, with increased body vibration
Third stage	2.98 g
Orbital coast	0

Accelerations measured by the robot’s own inertial module differ only slightly from those recorded by the spacecraft’s sensors [1], which is the validation that matters given that the robot carries its own load recording system for exactly this purpose [3]; the spikes in the record come from vibration of the robot’s body, because not all of its joints were fixed for launch and those joints had zero stiffness and transmitted vibration into the structure.

Ground operations

The flight article was reworked so it could be stowed in the descent module, given a space communications headset on the head module, and revised in its control system and individual components [1]. Qualification then covered fitting the robot into a purpose-built flight frame in the production shop; fitting on the Soyuz MS trainer mock-up; loading and unloading in a flight Soyuz MS descent module with stowage and fixation in the frame; the voice channel with the headset on a flight vehicle; vibration resistance; electromagnetic compatibility; a fire safety expert review; a sanitary and epidemiological expert review; a functionality check; and a check of operation from the autonomous power source. The link and module load recording system exists to close the loop on exactly these load cases [3].

Crew preparation was shaped by the fact that Soyuz MS-14 flew uncrewed. A. A. Skvortsov was trained as both operator 1, controlling the robot remotely, and operator 2, interacting with it directly, because A. N. Ovchinin was already aboard the ISS [1]. Skvortsov trained on a prototype at the Cosmonaut Training Centre, with the flight-standard system at RSC Energia, and with the flight master device and virtual robot models at Baikonur. The master device’s control glove was remade to the anthropometry of his hands.

Total crew time for the experiment was 57 hours 30 minutes: 28 hours from Ovchinin, commander of ISS-59/60, and 29 hours 30 minutes from Skvortsov, flight engineer 7 [1]. Ground ergonomics work had put average continuous operator sessions at about one hour, which is the unit the on-orbit sessions were built from [4].

On orbit the two cosmonauts moved the system from the descent module to MIM-2 and back [1]. All planned sessions were completed: remote control for typical flight operations, meaning work with onboard tools and a simulated wiping of station panel surfaces; a video greeting to Earth; and voice dialog between the cosmonauts and the robot. Video, audio, camera stills, video from both ISS and robot cameras, and telemetry from all flight phases were returned.

The ergonomics program behind the operator interface was run on the ground with more than 30 participants, cosmonauts and qualified instructors, wearing the master device, a VR display helmet and medical state monitoring sensors [4]. Average continuous work in the virtual environment was about one hour per operator. The study did not confirm published claims of adverse visual, proprioceptive, vestibular or muscular reactions that would prevent task execution. The sense of presence was rated positive but lower than expected, attributed to the absence of haptic feedback, geometric distortions, deficiencies in the VR helmet and insufficient modeling of physical conditions.

Technologies developed

The experiment’s findings are published as a numbered list, and three of them are load-bearing.

Copying control survives weightlessness. During the teleoperation scenarios the robot was rigidly fixed in MIM-2, no change in its dynamic characteristics was observed in weightlessness, and the system sustained copying control with dynamic parameters remaining normal [1]. The cosmonaut’s ability to drive it remotely in copying mode using the exoskeleton and virtual reality helmet, while himself in an unsupported environment, was confirmed. The qualifier is the useful part: to make control stable it was found advisable to provide special means of restraining the operating cosmonaut, since an unrestrained operator pushes himself away from the motion he is trying to command.

The structure survives the ride, against a design intent of up to five launches to orbit [3]. The system retained functionality under the external factors of ascent and descent, and autonomous tests after return confirmed serviceability and structural reliability [1]. Over the 16-day flight no noticeable influence of cosmic radiation or electromagnetic effects on the system’s functioning was detected.

The voice interface worked. The dialog experiment demonstrated the usability of a voice interface for intravehicular activity aboard an orbital station [1], one of the several command modalities the program has developed alongside gesture control, augmented-reality immersion and leader-follower operation [6].

Against these, the program’s own conclusion is that active use of anthropomorphic robots aboard orbital stations will require modification of the robot’s systems and possibly of the station infrastructure [1]. Further development is aimed at supporting cosmonauts in open space and in a prospective habitation system including a visited lunar orbital station and a planetary base, by raising functionality and autonomy through artificial intelligence.

References