Skip to content

Research · Compiled from public sources

The Hardware Behind Robot Training Data: Headsets, Capture Glasses, Grippers, Leader Arms and Suits

42 devices used to make robot training data (VR headsets, first-person cameras, handheld grippers, leader arms, motion-capture suits and gloves): what each costs, what it is used for and who uses it; 73 facts, each linked to its source.

Covers
23 Apr 2023 to 18 Oct 2026
Dated by when each event occurred
Calculated
18 Oct 2026, 20:07 UTC
Sources last checked 18 Oct 2026
Where it comes from
Makers' and labs' pages, papers and repositories, stores and reporting by major outlets, each linked with a verbatim quote kept in the dataset.
Method
Revision 1. Terms defined
CC BY 4.0, citing this report
Every figure as CSVJSON
devices
42
sourced facts
73

What the data shows

  1. 1

    Skill Capture Glove is the most documented of the 42 devices (5 sourced facts).

    Figure 1
  2. 2

    15 list prices on record, from open-source arm kits to motion-capture suits.

    Figure 2
  3. 3

    The most recent fact on record (14 Oct 2026): NPR reported that textile factory workers in Karur, India, are handed GoPros and smart glasses to record their work as egocentric robot training data; Objectways, an Indian tech company in Karur, gathers such data for AI developers.

    Figure 5

Figure 1

The hardware people use to make robot training data

Which headsets, cameras, grippers, arms and suits are used to collect robot training data, what do they cost and who uses them?

DeviceUsed forSourced facts
Skill Capture GloveSundaySunday converts Skill Capture Glove recordings into equivalent robot training data with a 90% success rate.
5
Apple Vision ProApple Newsroomn/a
4
Tesla Optimus data collection rigBusiness InsiderBusiness Insider reported that Tesla data collectors are sometimes outfitted with haptic gloves that track fine hand movements.
4
Universal Manipulation Interface (UMI)Pantheon Researchn/a
4
GELLOGELLO project (UC Berkeley)n/a
3
Meta Quest 3Meta Newsroomn/a
3
Mobile ALOHAMIT Technology Reviewn/a
3
Project Aria Gen 2Project Aria Docs (Meta)n/a
3
Ray-Ban Meta smart glassesMeta Newsroomn/a
3
ALOHAALOHA projectn/a
2
ALOHA 2Google DeepMindn/a
2
DexUMIarXivDexUMI uses a wearable hand exoskeleton, built for the XHand and Inspire Hand robot hands, to collect dexterous manipulation data with the human hand as the interface.
2
MANUS glovesMANUSn/a
2
Open-TeleVisionGitHub (OpenTeleVision)The Open-TeleVision teleoperation code works with both Apple Vision Pro and Meta Quest 3 headsets.
2
Project Aria glassesarXivn/a
2
SO-100Hugging Facen/a
2
SO-101GitHub (The Robot Studio)n/a
2
ACEarXivACE is a cross-platform teleoperation system that pairs a hand-facing camera with an exoskeleton on a portable base to capture finger and wrist poses for dexterous robot hands and grippers.
1
AgiBot World teleoperation systemsarXivAgiBot collected its AgiBot World data with two teleoperation systems: VR headset control and whole-body motion capture control.
1
AirExoAirExo projectn/a
1
Apple Vision Pro (M5)Apple Newsroomn/a
1
Build AI Gen 1Hugging Face (Build AI)n/a
1
DexCaparXivn/a
1
GoPro (head-mounted)NPRn/a
1
Headband phone mountLos Angeles Timesn/a
1
HOMIEarXivn/a
1
Koch low-cost robot armGitHub (Alexander Koch)n/a
1
MANUS Metagloves ProMANUSn/a
1
Oculus Quest 2DROID projectn/a
1
Project AriaGitHub (facebookresearch)n/a
1
Rokoko glovesarXivn/a
1
Rokoko Smartgloves IIRokokon/a
1
Sensei data collection platformY Combinatorn/a
1
SO-100 and SO-101GitHub (The Robot Studio)n/a
1
Stationary AI (ALOHA kit)Trossen Roboticsn/a
1
The Stick (Dobb-E)Dobb-E projectn/a
1
TWIST2arXivTWIST2, a humanoid data collection system, uses a PICO 4 Ultra (PICO4U) VR headset for real-time whole-body human motion, plus a custom 2-DoF robot neck costing about $250 for egocentric vision.
1
xr_teleoperateGitHub (Unitree Robotics)Unitree's xr_teleoperate repository teleoperates Unitree humanoid robots from XR headsets such as Apple Vision Pro, PICO 4 Ultra Enterprise or Meta Quest 3.
1
XRoboToolkitarXivByteDance's PICO team released XRoboToolkit, an OpenXR-based robot teleoperation framework that supports PICO 4 Ultra and Meta Quest 3 headsets.
1
Xsens Link suitXsens (Movella)n/a
1
Xsens motion captureXsens (Movella)n/a
1
Xsens motion capture suitXsens (Movella)Xsens describes humanoid robot training in which an operator wearing an Xsens suit teleoperates the robot in real time while the motion data is captured for AI training.
1

Base: 42 devices, 73 facts, each linked to the page that states it (reviewed 18 Oct 2026).

Prices are list prices as stated by the maker or reporting.

Credit: compiled by RobotTraining.jobs Research, The Hardware Behind Robot Training Data: Headsets, Capture Glasses, Grippers, Leader Arms and Suits, checked 18 Oct 2026.

CSV

Figure 2

Prices

What does each piece of data-collection hardware cost?

DevicePrice as stated
Koch low-cost robot armGitHub (Alexander Koch)Alexander Koch's open low-cost robot arm costs about $250, and a second leader arm for teleoperation about $180, for a total of $430.
SO-101GitHub (The Robot Studio)The SO-101 bill of materials for a leader and follower arm pair totals $229.88 at the listed US part prices.
Stationary AI (ALOHA kit)Trossen RoboticsTrossen Robotics lists its Aloha Stationary AI bimanual data collection kit at $23,995.95.
Skill Capture GloveBusiness InsiderSunday CEO Tony Zhao said the Skill Capture Glove gives two orders of magnitude better capital efficiency than teleoperation: $200 versus $20,000.
HOMIEarXivHOMIE, a humanoid teleoperation cockpit combining a pedal, an isomorphic exoskeleton arm and motion-sensing gloves, costs about $500.
Sensei data collection platformY CombinatorYC startup Sensei said its sensorized exoskeleton arm platform for collecting robot training data costs less than $300, against $40,000+ for a teleop setup with a physical robot.
DexCaparXivStanford researchers kept the total cost of the DexCap wearable hand motion capture rig within a $4,000 budget.
Mobile ALOHAMIT Technology ReviewMIT Technology Review reported that Stanford researchers built Mobile ALOHA for $32,000 using off-the-shelf robot parts and 3D-printed hardware.
Apple Vision ProApple NewsroomApple Vision Pro launched in the US at a starting price of $3,499 with 256GB of storage.
The Stick (Dobb-E)Dobb-E projectNYU's Dobb-E team built its demonstration tool, the Stick, from a $25 reacher-grabber, 3D printed parts and an iPhone.
Meta Quest 3Meta NewsroomMeta Quest 3 launched at $499.99 for the 128GB version and $649.99 for 512GB.
Ray-Ban Meta smart glassesMeta NewsroomThe second-generation Ray-Ban Meta smart glasses launched at a starting price of $299.
AirExoAirExo projectAirExo, a dual-arm exoskeleton for teleoperation and demonstration collection, costs approximately $300 per arm.
GELLOGELLO project (UC Berkeley)The parts for a GELLO leader-arm teleoperation device cost less than $300.
ALOHAALOHA projectThe original ALOHA low-cost open-source bimanual teleoperation system was built on a $20k budget.

Base: 15 facts from 15 devices.

Nothing is converted between currencies.

Credit: compiled by RobotTraining.jobs Research, The Hardware Behind Robot Training Data: Headsets, Capture Glasses, Grippers, Leader Arms and Suits, checked 18 Oct 2026.

CSV

Figure 3

Who uses it

Which labs and companies use each device to collect data?

DeviceAs stated
Open-TeleVisionGitHub (OpenTeleVision)The Open-TeleVision teleoperation code works with both Apple Vision Pro and Meta Quest 3 headsets.
Skill Capture GloveSundaySunday says it has shipped thousands of Skill Capture Gloves to its paid data collectors, whom it calls Memory Developers.
xr_teleoperateGitHub (Unitree Robotics)Unitree's xr_teleoperate repository teleoperates Unitree humanoid robots from XR headsets such as Apple Vision Pro, PICO 4 Ultra Enterprise or Meta Quest 3.
Xsens motion capture suitXsens (Movella)Xsens describes humanoid robot training in which an operator wearing an Xsens suit teleoperates the robot in real time while the motion data is captured for AI training.
GoPro (head-mounted)NPRNPR reported that textile factory workers in Karur, India, are handed GoPros and smart glasses to record their work as egocentric robot training data; Objectways, an Indian tech company in Karur, gathers such data for AI developers.
Universal Manipulation Interface (UMI)Pantheon ResearchPantheon built its own in-house UMI hardware for a data collection operation that it says reached over 1 million tasks in eight weeks at $10 per hour of data.
Headband phone mountLos Angeles TimesThe Los Angeles Times reported that Instawork hands out headbands with phone mounts so people in Los Angeles can film their chores at home as robot training data.
MANUS glovesMANUSIn NVIDIA's EgoScale research, operators wore MANUS gloves to capture 25 joint transforms per hand, with Vive trackers for the wrist, to align egocentric human video with robot control.
Xsens motion captureXsens (Movella)Chinese humanoid maker Ti5 Robotics uses Xsens inertial motion capture, which Xsens says sped up its humanoid motion modeling by 30%.
MANUS glovesMANUSMANUS says NVIDIA made Isaac Teleop generally available at GTC 2026 with MANUS gloves as the official data glove, integrated through a dedicated plugin.
GELLOBusiness InsiderBusiness Insider reported that OpenAI's robotics lab has data collectors using 3D-printed GELLO controllers to operate two Franka robots.
Skill Capture GloveSundaySunday converts Skill Capture Glove recordings into equivalent robot training data with a 90% success rate.
TWIST2arXivTWIST2, a humanoid data collection system, uses a PICO 4 Ultra (PICO4U) VR headset for real-time whole-body human motion, plus a custom 2-DoF robot neck costing about $250 for egocentric vision.
Tesla Optimus data collection rigBusiness InsiderBusiness Insider reported that Tesla data collectors are sometimes outfitted with haptic gloves that track fine hand movements.
Ray-Ban Meta smart glassesBusiness InsiderBusiness Insider reported that people paid to film chores for robot training wear Meta's Ray-Ban glasses; micro1's CEO cited rates of $25 to $50 an hour for this work.
Tesla Optimus data collection rigBusiness InsiderBusiness Insider reported that from late June 2025 Tesla moved Optimus data collection from motion-capture suits and teleoperation to workers recording themselves with five in-house cameras on a helmet and heavy backpack.
XRoboToolkitarXivByteDance's PICO team released XRoboToolkit, an OpenXR-based robot teleoperation framework that supports PICO 4 Ultra and Meta Quest 3 headsets.
SO-100Hugging FaceHugging Face pretrained SmolVLA on 487 community-shared datasets recorded on the SO-100 arm.
DexUMIarXivDexUMI uses a wearable hand exoskeleton, built for the XHand and Inspire Hand robot hands, to collect dexterous manipulation data with the human hand as the interface.
Project Aria glassesarXivNew York University's EgoZero learns robot manipulation policies from human demonstrations recorded on Project Aria smart glasses, with zero robot data.
Apple Vision ProarXivApple researchers used Apple Vision Pro to record EgoDex, 829 hours of egocentric video with paired 3D hand and finger tracking, released publicly for dexterous manipulation research.
ALOHA 2Google DeepMindGoogle DeepMind trained its Gemini Robotics model primarily on data from the ALOHA 2 bi-arm platform.
AgiBot World teleoperation systemsarXivAgiBot collected its AgiBot World data with two teleoperation systems: VR headset control and whole-body motion capture control.
Project Aria glassesEgoMimic project (Georgia Tech)Georgia Tech's EgoMimic framework captures human embodiment data (egocentric video with 3D hand tracking) on Project Aria glasses and co-trains robot policies on human and robot data.
ACEarXivACE is a cross-platform teleoperation system that pairs a hand-facing camera with an exoskeleton on a portable base to capture finger and wrist poses for dexterous robot hands and grippers.
Tesla Optimus data collection rigFortuneTesla's Optimus Data Collection Operators wore motion-capture suits and VR headsets for more than seven hours a day, for $25.25 to $48 an hour.
Apple Vision ProNVIDIA NewsroomNVIDIA's AI and simulation teleoperation reference workflow starts with developers capturing a small number of teleoperated demonstrations on Apple Vision Pro, which the MimicGen NIM then expands into synthetic motion data.
Oculus Quest 2DROID projectThe DROID dataset (76k trajectories, 350 hours) was collected on a standard rig of a Franka Panda arm, Zed stereo cameras and an Oculus Quest 2 headset with controllers for teleoperation.
Meta Quest 3arXivOPEN TEACH, an open-source teleoperation system from New York University and Meta researchers, runs on the Meta Quest 3, which its paper prices at $500, and controls Franka, xArm, Jaco and Allegro robots.
Rokoko glovesarXivStanford's DexCap portable hand motion capture system tracks finger motion with Rokoko motion capture gloves.

Base: 30 facts from 24 devices.

Credit: compiled by RobotTraining.jobs Research, The Hardware Behind Robot Training Data: Headsets, Capture Glasses, Grippers, Leader Arms and Suits, checked 18 Oct 2026.

CSV

Figure 4

Open-source designs

Which devices can anyone build from open designs?

DeviceAs stated
DexUMIGitHub (real-stanford)DexUMI publishes CAD models and an assembly tutorial for its Inspire Hand and XHand exoskeletons.
Project AriaGitHub (facebookresearch)Meta's projectaria_tools, the C++ and Python toolkit for working with Project Aria recordings, is open source under the Apache-2.0 licence.
SO-100 and SO-101GitHub (The Robot Studio)The SO-100 and SO-101 arm designs are published by The Robot Studio on GitHub under the Apache-2.0 licence and built to work with Hugging Face's open-source LeRobot library.
Open-TeleVisionOpen-TeleVision projectOpen-TeleVision, an immersive stereo-vision teleoperation system used to collect humanoid robot training data, was released as open source in July 2024.
Universal Manipulation Interface (UMI)arXivUMI's hardware and software were open-sourced; the code repository is released under the MIT license.
ALOHA 2arXivThe ALOHA 2 team open-sourced all ALOHA 2 hardware designs, a detailed tutorial and a MuJoCo simulation model.
Mobile ALOHANew AtlasMobile ALOHA is fully open source, with a published parts list and build guide.
GELLOarXivGELLO, a 3D-printed leader-arm teleoperation framework from UC Berkeley, was released with all software and hardware open-sourced.
ALOHAALOHA projectALOHA was released as a low-cost open-source hardware system for bimanual teleoperation, with kits sold by Trossen Robotics.

Base: 9 facts from 9 devices.

Credit: compiled by RobotTraining.jobs Research, The Hardware Behind Robot Training Data: Headsets, Capture Glasses, Grippers, Leader Arms and Suits, checked 18 Oct 2026.

CSV

Figure 5

Every sourced fact

What exactly has been published about each device?

DeviceKindFact
ACEarXivUse in data collectionACE is a cross-platform teleoperation system that pairs a hand-facing camera with an exoskeleton on a portable base to capture finger and wrist poses for dexterous robot hands and grippers.
AgiBot World teleoperation systemsarXivUse in data collectionAgiBot collected its AgiBot World data with two teleoperation systems: VR headset control and whole-body motion capture control.
AirExoAirExo projectPriceAirExo, a dual-arm exoskeleton for teleoperation and demonstration collection, costs approximately $300 per arm.
ALOHAALOHA projectPriceThe original ALOHA low-cost open-source bimanual teleoperation system was built on a $20k budget.
ALOHAALOHA projectOpen sourceALOHA was released as a low-cost open-source hardware system for bimanual teleoperation, with kits sold by Trossen Robotics.
ALOHA 2Google DeepMindWho uses itGoogle DeepMind trained its Gemini Robotics model primarily on data from the ALOHA 2 bi-arm platform.
ALOHA 2arXivOpen sourceThe ALOHA 2 team open-sourced all ALOHA 2 hardware designs, a detailed tutorial and a MuJoCo simulation model.
Apple Vision ProarXivWho uses itApple researchers used Apple Vision Pro to record EgoDex, 829 hours of egocentric video with paired 3D hand and finger tracking, released publicly for dexterous manipulation research.
Apple Vision ProNVIDIA NewsroomWho uses itNVIDIA's AI and simulation teleoperation reference workflow starts with developers capturing a small number of teleoperated demonstrations on Apple Vision Pro, which the MimicGen NIM then expands into synthetic motion data.
Apple Vision ProApple NewsroomReleaseApple Vision Pro went on sale in the US on Friday, February 2, 2024, with pre-orders from January 19.
Apple Vision ProApple NewsroomPriceApple Vision Pro launched in the US at a starting price of $3,499 with 256GB of storage.
Apple Vision Pro (M5)Apple NewsroomReleaseApple released an updated Vision Pro with the M5 chip and Dual Knit Band on October 22, 2025, still starting at $3,499.
Build AI Gen 1Hugging Face (Build AI)SpecsBuild AI's Egocentric-10K factory dataset was recorded on its own Build AI Gen 1 monocular head-mounted camera at 1080p and 30 fps with a 128 by 67 degree field of view.
DexCaparXivPriceStanford researchers kept the total cost of the DexCap wearable hand motion capture rig within a $4,000 budget.
DexUMIGitHub (real-stanford)Open sourceDexUMI publishes CAD models and an assembly tutorial for its Inspire Hand and XHand exoskeletons.
DexUMIarXivUse in data collectionDexUMI uses a wearable hand exoskeleton, built for the XHand and Inspire Hand robot hands, to collect dexterous manipulation data with the human hand as the interface.
GELLOBusiness InsiderWho uses itBusiness Insider reported that OpenAI's robotics lab has data collectors using 3D-printed GELLO controllers to operate two Franka robots.
GELLOGELLO project (UC Berkeley)PriceThe parts for a GELLO leader-arm teleoperation device cost less than $300.
GELLOarXivOpen sourceGELLO, a 3D-printed leader-arm teleoperation framework from UC Berkeley, was released with all software and hardware open-sourced.
GoPro (head-mounted)NPRWho uses itNPR reported that textile factory workers in Karur, India, are handed GoPros and smart glasses to record their work as egocentric robot training data; Objectways, an Indian tech company in Karur, gathers such data for AI developers.
Headband phone mountLos Angeles TimesWho uses itThe Los Angeles Times reported that Instawork hands out headbands with phone mounts so people in Los Angeles can film their chores at home as robot training data.
HOMIEarXivPriceHOMIE, a humanoid teleoperation cockpit combining a pedal, an isomorphic exoskeleton arm and motion-sensing gloves, costs about $500.
Koch low-cost robot armGitHub (Alexander Koch)PriceAlexander Koch's open low-cost robot arm costs about $250, and a second leader arm for teleoperation about $180, for a total of $430.
MANUS glovesMANUSWho uses itIn NVIDIA's EgoScale research, operators wore MANUS gloves to capture 25 joint transforms per hand, with Vive trackers for the wrist, to align egocentric human video with robot control.
MANUS glovesMANUSWho uses itMANUS says NVIDIA made Isaac Teleop generally available at GTC 2026 with MANUS gloves as the official data glove, integrated through a dedicated plugin.
MANUS Metagloves ProMANUSSpecsMANUS Metagloves Pro use electromagnetic field tracking and capture 25 degrees of freedom per hand.
Meta Quest 3arXivWho uses itOPEN TEACH, an open-source teleoperation system from New York University and Meta researchers, runs on the Meta Quest 3, which its paper prices at $500, and controls Franka, xArm, Jaco and Allegro robots.
Meta Quest 3Meta NewsroomReleaseMeta Quest 3 reached stores on October 10, 2023, after pre-orders opened at Meta Connect.
Meta Quest 3Meta NewsroomPriceMeta Quest 3 launched at $499.99 for the 128GB version and $649.99 for 512GB.
Mobile ALOHAMIT Technology ReviewPriceMIT Technology Review reported that Stanford researchers built Mobile ALOHA for $32,000 using off-the-shelf robot parts and 3D-printed hardware.
Mobile ALOHANew AtlasOpen sourceMobile ALOHA is fully open source, with a published parts list and build guide.
Mobile ALOHAarXivReleaseMobile ALOHA, a low-cost whole-body teleoperation system for collecting bimanual mobile manipulation data, was published in January 2024.
Oculus Quest 2DROID projectWho uses itThe DROID dataset (76k trajectories, 350 hours) was collected on a standard rig of a Franka Panda arm, Zed stereo cameras and an Oculus Quest 2 headset with controllers for teleoperation.
Open-TeleVisionGitHub (OpenTeleVision)Use in data collectionThe Open-TeleVision teleoperation code works with both Apple Vision Pro and Meta Quest 3 headsets.
Open-TeleVisionOpen-TeleVision projectOpen sourceOpen-TeleVision, an immersive stereo-vision teleoperation system used to collect humanoid robot training data, was released as open source in July 2024.
Project AriaGitHub (facebookresearch)Open sourceMeta's projectaria_tools, the C++ and Python toolkit for working with Project Aria recordings, is open source under the Apache-2.0 licence.
Project Aria Gen 2Project Aria Docs (Meta)SpecsAria Gen 2's RGB camera is a 12 megapixel sensor with a 133 by 99 degree field of view that captures full resolution (4032 by 3024) at 24 fps.
Project Aria Gen 2Project Aria Docs (Meta)SpecsAria Gen 2 runs hand tracking on the device at 30 Hz with 21 joint landmarks per hand, alongside eye gaze tracking at up to 90 Hz.
Project Aria Gen 2CNBCReleaseMeta unveiled Aria Gen 2 research glasses for AI, robotics and machine perception research, to be made available to academic and commercial research labs, without a wider availability date.
Project Aria glassesarXivWho uses itNew York University's EgoZero learns robot manipulation policies from human demonstrations recorded on Project Aria smart glasses, with zero robot data.
Project Aria glassesEgoMimic project (Georgia Tech)Who uses itGeorgia Tech's EgoMimic framework captures human embodiment data (egocentric video with 3D hand tracking) on Project Aria glasses and co-trains robot policies on human and robot data.
Ray-Ban Meta smart glassesBusiness InsiderWho uses itBusiness Insider reported that people paid to film chores for robot training wear Meta's Ray-Ban glasses; micro1's CEO cited rates of $25 to $50 an hour for this work.
Ray-Ban Meta smart glassesMeta NewsroomPriceThe second-generation Ray-Ban Meta smart glasses launched at a starting price of $299.
Ray-Ban Meta smart glassesMeta NewsroomSpecsRay-Ban Meta glasses have an ultra-wide 12 MP camera that records 1080p video clips of up to 60 seconds.
Rokoko glovesarXivWho uses itStanford's DexCap portable hand motion capture system tracks finger motion with Rokoko motion capture gloves.
Rokoko Smartgloves IIRokokoSpecsRokoko Smartgloves II sample their internal IMUs at 250 Hz.
Sensei data collection platformY CombinatorPriceYC startup Sensei said its sensorized exoskeleton arm platform for collecting robot training data costs less than $300, against $40,000+ for a teleop setup with a physical robot.
Skill Capture GloveSundaySpecsSunday's glove records motion through force sensors, cameras and joint potentiometers, and Memo's hand is designed to mirror the glove's shape and sensors.
Skill Capture GloveSundayWho uses itSunday says it has shipped thousands of Skill Capture Gloves to its paid data collectors, whom it calls Memory Developers.
Skill Capture GloveBusiness InsiderPriceSunday CEO Tony Zhao said the Skill Capture Glove gives two orders of magnitude better capital efficiency than teleoperation: $200 versus $20,000.
Skill Capture GloveSundayReleaseSunday unveiled the Skill Capture Glove with its Memo robot and ACT-1 model, which it says was trained without a single trajectory of teleoperation data.
Skill Capture GloveSundayUse in data collectionSunday converts Skill Capture Glove recordings into equivalent robot training data with a 90% success rate.
SO-100Hugging FaceWho uses itHugging Face pretrained SmolVLA on 487 community-shared datasets recorded on the SO-100 arm.
SO-100TechCrunchReleaseHugging Face's LeRobot team and The Robot Studio debuted the SO-100 arm for around $100 in October 2024.
SO-100 and SO-101GitHub (The Robot Studio)Open sourceThe SO-100 and SO-101 arm designs are published by The Robot Studio on GitHub under the Apache-2.0 licence and built to work with Hugging Face's open-source LeRobot library.
SO-101GitHub (The Robot Studio)PriceThe SO-101 bill of materials for a leader and follower arm pair totals $229.88 at the listed US part prices.
SO-101TechCrunchReleaseHugging Face released the SO-101 arm, the faster-to-assemble successor to the SO-100, starting at $100, with supplier prices ranging up to around $500.
Stationary AI (ALOHA kit)Trossen RoboticsPriceTrossen Robotics lists its Aloha Stationary AI bimanual data collection kit at $23,995.95.
Tesla Optimus data collection rigBusiness InsiderSpecsBusiness Insider reported that Tesla's camera rig backpack weighs 30 to 40 pounds and is wired to a headset that can feed workers AI-generated prompts.
Tesla Optimus data collection rigBusiness InsiderUse in data collectionBusiness Insider reported that Tesla data collectors are sometimes outfitted with haptic gloves that track fine hand movements.
Tesla Optimus data collection rigBusiness InsiderUse in data collectionBusiness Insider reported that from late June 2025 Tesla moved Optimus data collection from motion-capture suits and teleoperation to workers recording themselves with five in-house cameras on a helmet and heavy backpack.
Tesla Optimus data collection rigFortuneUse in data collectionTesla's Optimus Data Collection Operators wore motion-capture suits and VR headsets for more than seven hours a day, for $25.25 to $48 an hour.
The Stick (Dobb-E)Dobb-E projectPriceNYU's Dobb-E team built its demonstration tool, the Stick, from a $25 reacher-grabber, 3D printed parts and an iPhone.
TWIST2arXivUse in data collectionTWIST2, a humanoid data collection system, uses a PICO 4 Ultra (PICO4U) VR headset for real-time whole-body human motion, plus a custom 2-DoF robot neck costing about $250 for egocentric vision.
Universal Manipulation Interface (UMI)Pantheon ResearchWho uses itPantheon built its own in-house UMI hardware for a data collection operation that it says reached over 1 million tasks in eight weeks at $10 per hour of data.
Universal Manipulation Interface (UMI)arXivReleaseUMI, a hand-held gripper system for collecting robot demonstrations in the wild without a robot, was published by Stanford, Columbia and Toyota Research Institute researchers.
Universal Manipulation Interface (UMI)UMI projectSpecsThe UMI data collection device is a hand-held parallel jaw gripper with a mounted GoPro camera that records a 155 degree fisheye view.
Universal Manipulation Interface (UMI)arXivOpen sourceUMI's hardware and software were open-sourced; the code repository is released under the MIT license.
xr_teleoperateGitHub (Unitree Robotics)Use in data collectionUnitree's xr_teleoperate repository teleoperates Unitree humanoid robots from XR headsets such as Apple Vision Pro, PICO 4 Ultra Enterprise or Meta Quest 3.
XRoboToolkitarXivUse in data collectionByteDance's PICO team released XRoboToolkit, an OpenXR-based robot teleoperation framework that supports PICO 4 Ultra and Meta Quest 3 headsets.
Xsens Link suitXsens (Movella)SpecsThe Xsens Link full-body suit carries 17 embedded sensors (plus one extra), updates at up to 240 Hz and runs about 3 hours per swappable battery.
Xsens motion captureXsens (Movella)Who uses itChinese humanoid maker Ti5 Robotics uses Xsens inertial motion capture, which Xsens says sped up its humanoid motion modeling by 30%.
Xsens motion capture suitXsens (Movella)Use in data collectionXsens describes humanoid robot training in which an operator wearing an Xsens suit teleoperates the robot in real time while the motion data is captured for AI training.

Base: All 73 facts. "Reported" means an outlet reports it from sources the company has not confirmed.

Amounts stay in the currency and form their source gives; nothing is converted or totalled.

Credit: compiled by RobotTraining.jobs Research, The Hardware Behind Robot Training Data: Headsets, Capture Glasses, Grippers, Leader Arms and Suits, checked 18 Oct 2026.

CSV

Robotics labs · Fleet operators · Data vendors · Research teams

Buy the data behind these reports

employer job boards read every day
103employer job boards read every day
every role's history kept since
12 Sep 2026every role's history kept since
apply clicks in the last 30 days
392apply clicks in the last 30 days

  • The raw records

    Every robot training role we track: title, employer, place, kind of work, pay as stated, and the days it was first and last open.

  • A regular feed

    New and closed roles every week or month, as file drops or through the data endpoints.

  • A custom cut

    Any figure here by country, city, kind of work, vehicle, employer or date, or your own hiring benchmarked against the market.

Each report's own figures stay free to download and reuse under CC BY 4.0.

How it was measured

Makers' and labs' pages, papers and repositories, stores and reporting by major outlets, each linked with a verbatim quote kept in the dataset. Each row carries a link to its publication and gives the number exactly as published, with no estimates or conversions. A note under every figure says what it covers.

Leader-follower arm
A hand-moved arm (the leader) whose motion a robot arm (the follower) copies, used to record demonstrations.
UMI
Universal Manipulation Interface: a handheld gripper with a camera for recording demonstrations without a robot.

Limits of this data

  • Built only from public sources. Anything never announced or reported is absent, and private-company figures are whatever the company or the press said, unaudited.
  • Store prices are as fetched and change.
  • Several devices have no published price; they are listed without one.

Collection and classification of listings are explained on the methodology page. Freshness, completeness and consistency are tested daily in the data quality report.

Monthly editions

The live page shows the dataset as of its latest review, and new events join it once they are made public. Every finished month also gets its own edition, locked two days after month end, so a number quoted from it stays fixed.

Quoting this report

You may quote, chart or republish these figures under CC BY 4.0, provided you link back to this page. The live numbers shift every day. For a figure that stays put, quote a monthly edition, or note the date you accessed this page.

Citation

RobotTraining.jobs (2026). The Hardware Behind Robot Training Data: Headsets, Capture Glasses, Grippers, Leader Arms and Suits (live data, 23 Apr 2023 to 18 Oct 2026). https://robottraining.jobs/research/robot-data-capture-hardware (accessed 18 Oct 2026)

BibTeX

@misc{robottrainingjobs_robot_data_capture_hardware_live,
  author = {{RobotTraining.jobs}},
  title = {The Hardware Behind Robot Training Data: Headsets, Capture Glasses, Grippers, Leader Arms and Suits (live data, 23 Apr 2023 to 18 Oct 2026)},
  year = {2026},
  howpublished = {\url{https://robottraining.jobs/research/robot-data-capture-hardware}},
  note = {Accessed 18 Oct 2026},
}

Files: every figure in one CSV (a row per value) or as JSON. Each figure can also be downloaded separately. Other open data: datasets and API.

Keep reading

Browse all 80 reports