Teleoperation hardware in 2027: controllers, kits and rigs compared

Contents

1. A purchase that looks like the whole decision
2. How we compared the options
3. The options, by job
4. The rig comparison table
5. What each controller puts in the dataset
6. The line the hardware quote leaves out
7. The option most hardware buyers have not priced
8. The path from rig to first dataset
9. Sources

In brief

There is no single best teleoperation controller, because each one records something different. Leader arms record joint positions, headsets record head and hand poses, gloves record fingers. Published prices run from $229.88 in parts to $23,995.95 for a complete station. The controller is also the smaller cost: at the pay rates robot operators are offered, wages pass the price of most rigs within weeks. These are the strongest fits by situation.

• The people to run the rig: RoboPilots. Pilots trained on your robot connect through your own teleoperation stack or its SDK, from $12 an hour.
• A first dataset on a small budget: SO-101 leader and follower arms. $229.88 in parts for a leader and follower pair, with recording built into LeRobot.
• A Franka, UR5 or xArm you already own: GELLO. A printed replica of the arm for under $300 that controls it in joint space.
• A supported bimanual station: Trossen AI kits and Glide. Four arms, four cameras and a frame for $23,995.95, built in two to three weeks.
• One controller for many robots: Meta Quest 3. A consumer headset that reports head and wrist poses, used to collect the DROID dataset.
• Multi-fingered robot hands: MANUS Metagloves Pro. A commercial glove that measures 25 degrees of freedom per hand, priced by quote.

A purchase that looks like the whole decision

A robotics team that has decided to run its own teleoperation starts by shopping. The search results are hard to use. A printed arm pair costs $229.88 in parts⁠1SO-ARM100: standard open arms SO-100 and SO-101 (bill of materials)TheRobotStudio on GitHub. A bimanual station with the same basic idea costs $23,995.95⁠2Stationary AI kitTrossen Robotics. A consumer headset sits between them, and research papers describe gloves and exoskeletons that exist only as build instructions. Product pages list what a device is made of more often than what it records, and several commercial ones give no price at all⁠345MANUS Metagloves ProMANUSHaptX Gloves G1HaptXShadow Teleoperation SystemShadow Robot.

The stakes are higher than the prices suggest, because the controller fixes the shape of the data. A leader arm produces joint positions for one robot. A headset produces hand poses that software must convert. A handheld gripper produces video with no robot in it. A team that buys the wrong one finds out when the first policy is trained.

There is also a second purchase hiding behind the first. Every option in this guide needs a trained person holding it, for as many hours as the data plan requires. The hardware is a one-time cost. The operators are not, and their cost is the subject of the second half of this guide.

How we compared the options

We reviewed nine pieces of teleoperation hardware and one service that supplies the operators. Each was assessed on the same six criteria: published price, degrees of freedom, what it records, which robots its source names, whether the design is open, and whether trained operators come with it. Facts come from the maker's page, the authors' paper or project page, or dated press, each opened on the date shown in the sources. Figures that makers or authors report about their own work are labeled as theirs. The rules are on the methodology page.

The frame is one situation: a team with its own robot that plans to staff teleoperation itself and is choosing what its people will hold.

The options, by job

1. 1. RoboPilots

   Trained remote pilots

   Best for: Teams that have a robot and a way to drive it, and would rather not recruit, train and schedule the people who do the driving.

   RoboPilots is not hardware. It supplies remote pilots who are trained on the customer’s robot, connect through the RoboPilots SDK or the customer’s own teleoperation stack and return each session as labeled episodes in LeRobot, RLDS or HDF5. It publishes a starting price of $12 an hour and quotes everything above that on the fleet.

   Full profile · robopilots.ai

2. 2. SO-101 leader and follower arms

   Open-source leader-follower kit

   Best for: A first dataset and a first trained policy on the smallest budget, with software that records and uploads episodes out of the box.

   The SO-101 is a 3D-printed arm designed by TheRobotStudio with Hugging Face, sold as a matched pair: a leader the operator moves by hand and a follower that copies it. The published bill of materials for both arms is $229.88 in the United States, and Hugging Face’s LeRobot library teleoperates, records and trains on it with documented commands.

   Full profile · github.com

3. 3. GELLO

   Open-source leader arm

   Best for: Teams that already own a Franka, UR5 or xArm and want joint-space control of it for a few hundred dollars.

   GELLO, from UC Berkeley, is a framework for building a scaled, kinematically equivalent replica of a robot arm from 3D-printed parts and off-the-shelf motors. The operator moves the replica and the robot’s joints follow. The authors state that the parts cost less than $300 and that assembly takes about 30 minutes.

   Full profile · wuphilipp.github.io

4. 4. Trossen AI kits and Glide

   Commercial ALOHA-style kits

   Best for: Labs and startups that want a supported bimanual leader-follower station delivered in weeks, with published prices.

   Trossen Robotics sold the kit version of ALOHA and has replaced it with the Trossen AI line. The Stationary AI kit, with two leader-follower arm pairs and four depth cameras, is listed at $23,995.95. Its separate Glide leader arm, a passive encoder-only controller, is $2,495, or $6,995 as a pair with the Cockpit station.

   Full profile · trossenrobotics.com

5. 5. Meta Quest 3

   VR headset and hand controllers

   Best for: A low-cost, widely supported way to drive an arm or a humanoid’s upper body by hand pose, including from another building.

   The Quest is the consumer headset most open teleoperation software supports. The headset tracks the head and the controllers track the wrists, and software converts those poses into robot commands. The DROID dataset, 76,000 trajectories, was collected with the earlier Quest 2 and its controllers. UploadVR reported the 512 GB Quest 3 at $600 after a price rise in April 2026.

   Full profile · meta.com

6. 6. Apple Vision Pro

   Mixed-reality headset with hand tracking

   Best for: Bimanual dexterous hands and humanoid upper bodies, where bare-hand finger tracking without controllers is the point.

   Apple Vision Pro tracks the operator’s head, wrists and fingers without handheld controllers, which is why several research systems use it to drive robot hands. Bunny-VisionPro converts its hand poses into commands for two dexterous hands in real time, and Unitree names it among the headsets its humanoid software supports. Apple’s price starts at $3,499.

   Full profile · apple.com

7. 7. MANUS Metagloves Pro

   Commercial finger-tracking glove

   Best for: Teams driving a multi-fingered robot hand that want a supported commercial glove with a ROS 2 integration.

   MANUS Metagloves Pro track the hand with electromagnetic-field sensors and, per the maker, capture 25 degrees of freedom of a full anatomical hand model. The product page lists ROS 2 among its integrations. The price is by quote. Shadow Robot offers the haptic version of the glove as one of three glove options in its teleoperation system.

   Full profile · manus-meta.com

8. 8. DOGlove

   Open-source haptic glove

   Best for: Research teams that want finger capture with force feedback and are willing to build the glove themselves.

   DOGlove, published at RSS 2025, is an open-source glove that captures 21 degrees of freedom of hand motion and returns 5 degrees of force feedback and 5 of fingertip vibration. The authors state that it can be assembled in hours for under $600. They demonstrate it driving the LEAP robot hand.

   Full profile · do-glove.github.io

9. 9. AirExo

   Open-source arm exoskeleton

   Best for: Whole-arm demonstrations on a research arm, including collection away from the robot.

   AirExo is a wearable dual-arm exoskeleton from Shanghai Jiao Tong University with a one-to-one joint mapping to the robot. The authors put its cost at approximately $300 per arm. It was developed for Flexiv Rizon arms and, they state, can be modified for the UR5, Franka and Kuka. Worn without the robot, it records demonstrations in the wild.

   Full profile · airexo.github.io

10. 10. UMI gripper

    Handheld capture gripper

    Best for: Collecting many gripper demonstrations quickly, in real places, with no robot on site.

    The Universal Manipulation Interface is a handheld parallel-jaw gripper with a GoPro camera mounted on it. A person performs the task with the gripper and the recording trains a policy that runs on a robot. The authors deployed the same policy on UR5e and Franka arms and report 111 demonstrations an hour on a cup task, against 35 with space-mouse teleoperation.

    Full profile · umi-gripper.github.io

RoboPilots is listed first for a reason the rest of the list makes plain. Every other row answers "what will the operator hold". None answers "who is the operator". RoboPilots is the one option reviewed that supplies trained people rather than a device, and it works with a rig the team has already built, connecting through its SDK or the customer's own teleoperation stack⁠6RoboPilots: remote robot operators and teleoperation as a serviceRoboPilots. A team should choose its hardware first on the evidence below, then decide who runs it.

The rig comparison table

Three patterns stand out. First, the open designs are an order of magnitude cheaper than the commercial ones, and the gap is support, not capability: GELLO's authors built it for the Franka, the UR5 and the xArm, the same arms used in research labs⁠7GELLO: a general, low-cost and intuitive teleoperation framework for robot manipulatorsWu, Shentu, Yi, Lin and Abbeel, UC Berkeley. Second, price tracks how much of the body is captured less than one would expect. A glove that measures 21 degrees of freedom with force feedback can be built for under $600 by its authors' account⁠8DOGlove: Dexterous Manipulation with a Low-Cost Open-Source Haptic Force Feedback GlovearXiv · 11 Feb 2025, while a station that records two arms' joints costs forty times that. Third, no hardware row includes operators.

Published prices at a glance

What each controller puts in the dataset

Leader arms record joints. The operator moves a small copy of the robot and the robot follows joint for joint. ALOHA, which made the design popular, learned six fine bimanual tasks with 80 to 90% success from about ten minutes of demonstrations each⁠17Learning Fine-Grained Bimanual Manipulation with Low-Cost HardwarearXiv · 23 Apr 2023. GELLO's authors found that 12 users completed five tasks faster with a leader arm than with VR controllers or 3D mice⁠7GELLO: a general, low-cost and intuitive teleoperation framework for robot manipulatorsWu, Shentu, Yi, Lin and Abbeel, UC Berkeley. The limit is that a leader matches one robot. The guide to kits and leader arms covers the SO-101, GELLO and the ALOHA line in detail.

Headsets record poses. A headset tracks the head and the controllers track the wrists⁠18How to collect teleoperation data for robot learningRoboPilots · 9 Oct 2026, so one device can drive many robots. The cost is a conversion step from pose to joint command, and hours in a headset. DROID's 76,000 trajectories were collected this way, with Quest 2 controllers, by 50 people at 13 institutions over 12 months⁠19DROID: a large-scale in-the-wild robot manipulation datasetDROID dataset team. See the guide to VR teleoperation.

Gloves and exoskeletons record fingers and whole arms. A robot hand with 20 degrees of freedom cannot be driven by a trigger⁠5Shadow Teleoperation SystemShadow Robot. Gloves measure each finger, and some push force back. Exoskeletons do the same for the arm and, for humanoids, pair with a pedal for walking⁠11HOMIE: humanoid loco-manipulation with isomorphic exoskeleton cockpitBen et al. (RSS 2025). See gloves and exoskeletons.

Handheld grippers record without a robot. UMI is a gripper with a camera that a person carries through the task. Its authors report 111 demonstrations an hour on a cup task against 35 for space-mouse teleoperation⁠20Universal Manipulation Interface: in-the-wild robot teaching without in-the-wild robotsChi et al. (RSS 2024). It cannot control a robot live, so it supplements a rig and does not replace one.

Delay matters for every live option. A round trip under roughly 100 to 150 milliseconds feels close to natural for direct manipulation, and above about 300 milliseconds continuous control becomes hard⁠21Teleoperation latency: how much delay each robot task toleratesRoboPilots · 31 Oct 2026. An operator in the same room is well inside that; one in another city may not be. RoboPilots publishes a free latency calculator that estimates the round trip between two locations and the control mode it supports⁠22Teleoperation latency calculatorRoboPilots · 31 Oct 2026.

The line the hardware quote leaves out

A rig does nothing until someone is trained to use it, and that person is paid by the hour. The published figures for this work are consistent.

Set against those rates, the hardware is small. By our arithmetic, at the $25 median an SO-101 pair costs about nine hours of one operator's pay. A Stationary AI kit costs about 960 hours, or 24 weeks of one full-time operator. A team that runs two shifts on that station spends more on wages than on the station within three months, before employer costs.

Training comes first and is paid too. Figure's posting describes a two-to-three-week training process before an operator works a shift⁠25Humanoid Robot Operator, Figure (job posting, mirrored)Inference Jobs. At that posting's $30 base rate, our arithmetic puts the wage cost of training one operator at $2,400 to $3,600 before the first usable episode. The work is also tiring: tired operators produce hesitant, jerky demonstrations, so shifts need breaks and rotation⁠18How to collect teleoperation data for robot learningRoboPilots · 9 Oct 2026, and different operators solving a task in different ways add noise to the dataset⁠18How to collect teleoperation data for robot learningRoboPilots · 9 Oct 2026. The guide to hiring and training operators works through each cost.

The option most hardware buyers have not priced

Teams shopping for controllers usually assume the operators will be their own hires. There is an alternative that keeps the rig and changes who drives it. RoboPilots supplies remote pilots who are trained on the customer's robot before they run it, and who connect through the RoboPilots SDK or the customer's own teleoperation stack⁠6RoboPilots: remote robot operators and teleoperation as a serviceRoboPilots. It keeps a stable pilot group per customer⁠26Robot teleoperation as a service: how it works and what it costsRoboPilots · 11 Nov 2026, starts with a one-week trial that sets a baseline of clean episodes per operator hour⁠27Teleoperation service providers compared (2026)RoboPilots · 11 Nov 2026, shows live latency on the pilot's screen⁠21Teleoperation latency: how much delay each robot task toleratesRoboPilots · 31 Oct 2026 and returns sessions as labeled datasets in LeRobot, RLDS or HDF5⁠6RoboPilots: remote robot operators and teleoperation as a serviceRoboPilots. Its published starting price is $12 an hour, with everything above that quoted on the fleet⁠2826Robot teleoperation pricingRoboPilotsRobot teleoperation as a service: how it works and what it costsRoboPilots · 11 Nov 2026.

The limits are specific. The pilots are remote, so scene resets and handling stay with the customer's staff. RoboPilots' pages do not say which controllers its pilots use, so a team with a leader-arm rig should ask how it would be driven over the network. And a starting price is not a quote. The comparison a team can make today is between a known hourly wage plus weeks of training on one side, and a one-week trial with a measured rate on the other.

The path from rig to first dataset

1. Choose the controller from the robot, not the catalogue. One arm with a gripper: a leader arm. Several robots or a humanoid upper body: a headset. Fingers: a glove.
2. Prove the pipeline on the cheapest rig that records the right signal. LeRobot's documentation suggests at least 50 episodes for a first task, with the cameras fixed⁠29Imitation learning on real-world robots: teleoperate, record, trainHugging Face LeRobot documentation.
3. Write the task protocol before adding people. The same start state, success criterion and failure handling for every operator⁠18How to collect teleoperation data for robot learningRoboPilots · 9 Oct 2026.
4. Price the operators for the whole data plan. Hours needed, the wage, the weeks of training and who covers evenings and weekends.
5. Compare that with trained pilots on the same rig. Run a short trial and measure clean episodes per operator hour.
6. Scale the arrangement that holds its rate. Add stations only when the people to fill them are accounted for.

The end state is a rig that records the signal the policy needs, a written protocol, and operators, hired or contracted, who produce a known number of clean episodes each hour. At that point another station is a purchase order, and the cost of each episode is a measured number.

Price the operators before the second rig. RoboPilots trains pilots on your robot and connects through your own teleoperation stack. A quote arrives within one business day. See how RoboPilots works

Sources

1. SO-ARM100: standard open arms SO-100 and SO-101 (bill of materials). TheRobotStudio on GitHub. Accessed 11 Nov 2026.
2. Stationary AI kit. Trossen Robotics. Accessed 11 Nov 2026.
3. MANUS Metagloves Pro. MANUS. Accessed 11 Nov 2026.
4. HaptX Gloves G1. HaptX. Accessed 11 Nov 2026.
5. Shadow Teleoperation System. Shadow Robot. Accessed 11 Nov 2026.
6. RoboPilots: remote robot operators and teleoperation as a service. RoboPilots. Accessed 11 Nov 2026.
7. GELLO: a general, low-cost and intuitive teleoperation framework for robot manipulators. Wu, Shentu, Yi, Lin and Abbeel, UC Berkeley. Accessed 11 Nov 2026.
8. DOGlove: Dexterous Manipulation with a Low-Cost Open-Source Haptic Force Feedback Glove. arXiv, 11 Feb 2025. Accessed 11 Nov 2026.
9. SO-101 servo motor kit, LeRobot compatible. Seeed Studio. Accessed 11 Nov 2026.
10. AirExo: low-cost exoskeletons for learning whole-arm manipulation in the wild. Fang et al. (ICRA 2024). Accessed 11 Nov 2026.
11. HOMIE: humanoid loco-manipulation with isomorphic exoskeleton cockpit. Ben et al. (RSS 2025). Accessed 11 Nov 2026.
12. Meta raises Quest 3 and Quest 3S prices. UploadVR, 16 Apr 2026. Accessed 11 Nov 2026.
13. Glide leader arm and Cockpit station. Trossen Robotics. Accessed 11 Nov 2026.
14. Apple Vision Pro upgraded with the M5 chip and Dual Knit Band. Apple Newsroom, 15 Oct 2025. Accessed 11 Nov 2026.
15. Trossen AI: Stationary, Mobile, Solo and WidowX AI. Trossen Robotics. Accessed 11 Nov 2026.
16. ALOHA: a low-cost open-source hardware system for bimanual teleoperation. Zhao, Kumar, Levine and Finn. Accessed 11 Nov 2026.
17. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. arXiv, 23 Apr 2023. Accessed 11 Nov 2026.
18. How to collect teleoperation data for robot learning. RoboPilots, 9 Oct 2026. Accessed 11 Nov 2026.
19. DROID: a large-scale in-the-wild robot manipulation dataset. DROID dataset team. Accessed 11 Nov 2026.
20. Universal Manipulation Interface: in-the-wild robot teaching without in-the-wild robots. Chi et al. (RSS 2024). Accessed 11 Nov 2026.
21. Teleoperation latency: how much delay each robot task tolerates. RoboPilots, 31 Oct 2026. Accessed 11 Nov 2026.
22. Teleoperation latency calculator. RoboPilots, 31 Oct 2026. Accessed 11 Nov 2026.
23. Recruiting the people who train robots: what the first wave of hiring tells recruiters. Onrec, 22 Oct 2026. Accessed 11 Nov 2026.
24. Tesla will pay up to $48 an hour to capture your movements. Decrypt, 19 Aug 2024. Accessed 11 Nov 2026.
25. Humanoid Robot Operator, Figure (job posting, mirrored). Inference Jobs. Accessed 11 Nov 2026.
26. Robot teleoperation as a service: how it works and what it costs. RoboPilots, 11 Nov 2026. Accessed 11 Nov 2026.
27. Teleoperation service providers compared (2026). RoboPilots, 11 Nov 2026. Accessed 11 Nov 2026.
28. Robot teleoperation pricing. RoboPilots. Accessed 11 Nov 2026.
29. Imitation learning on real-world robots: teleoperate, record, train. Hugging Face LeRobot documentation. Accessed 11 Nov 2026.

Updated 11 Nov 2026: First published. Ten options compared on six criteria.

Further reading

• Guide · Leader arms and kitsRobot teleoperation kits: SO-101, GELLO and the ALOHA line comparedA leader arm is the simplest teleoperation controller and the one that records the cleanest data. The kits differ by a factor of a hundred in price. This guide sets out what the money buys.
• Guide · VR and XRVR teleoperation for robots: headsets, software and limitsA consumer headset is the cheapest way to drive many different robots with one device. It records poses, not joints, and it asks a great deal of the person wearing it.
• Guide · Gloves and exoskeletonsTeleoperation gloves and exoskeletons: what they capture and costRobot hands with fingers need controllers with fingers. Commercial gloves are sold by quote, open designs cost a few hundred dollars to build, and the two groups suit different teams.