MUSCLE / MOTION
FMG arrays and an IMU sit on the forearm, pairing muscle-deformation-related contact pressure changes and inertial motion with egocentric video. With a head-mounted stereo camera plus armbands on both forearms, the hands stay free for natural contact, producing research data for gestures, hand pose and robot learning.
Data outputs, interface adaptation and delivery scope are confirmed per task.
Armbands on both forearms can work with head-mounted vision without covering the palms or fingertips. This is a capture approach for natural manipulation demonstrations; "glove-free" does not mean it replaces every tactile or hand-pose device.
The video shows a modular ring structure, inner contact units and an IMU. FMG deformation signals complement vision; forearm contact changes detected by the sensing units are not contact forces between fingers and objects.
Explore gesture recognition, 3D hand pose and fingertip pressure prediction, paired with Ego, UMI, glove or real-robot data. Whether it improves task performance requires independent labels, cross-person and cross-day tests and controlled robot-task comparisons.
FMG + IMU / IN THREE DIMENSIONS
Drag to view the ring exterior, inner contact surface and control area. Graphite black and champagne gold are material options; the display model is an independently simplified reconstruction.
SEE THE SIGNALS
From wearing position to what the data means: how FMG, IMU and the first-person view complement each other.
01 / CONTACT
Records contact changes related to forearm muscle deformation. The signal varies with placement, tightness and the individual, so wearing baselines and calibration data are kept.
This is not EMG, nor ground-truth contact force between fingertips and objects.
FMG / SEE THE SYSTEM
One wearable setup, two demos: visual observation and forearm signals in the same study.
01 / IN MOTION
Shows the head-mounted stereo camera with FMG armbands on both forearms, plus the 10×4 sensing array, IMU and 120 Hz stated in the video.
Download this demo02 / IN MOTION
Shows the modular ring structure, inner contact units and the two-arm wearing setup. Hand pose and fingertip pressure in the video are model-prediction targets, not direct measurements.
Download this demoSIGNAL ≠ PREDICTION ≠ ACTION
Raw capture, model estimates and robot control are managed separately, preserving data provenance and validation boundaries.
Channel outputs of the FMG sensing units, IMU, device timestamps, wearing baseline and quality flags. Range, units and channel mapping are confirmed per hardware version.
Captured dataExplore gesture, 3D hand pose and fingertip pressure prediction, keeping model version, confidence and reference labels. Predictions are not treated as direct ground-truth measurements.
Research direction · to be validatedPaired with vision, tactile or real-robot records to evaluate whether manipulation improves. Mapping human signals to robot actions still requires mapping, constraints and real-robot safety checks.
Task evaluationEGO + FMG + IMU
A head-mounted stereo camera records the viewpoint, and left and right armbands record signals from each side. Each session saves device IDs, side, placement and cross-device time mapping. Whether to add gloves, tactile end effectors or independent force sensors depends on the reference ground truth the task needs.
THE DETAILS MATTER
Forearm FMG array and IMU capture, pairable with Ego head-mounted vision.
| Parameter | Spec / configuration |
|---|---|
| Signal combination | FMG + IMU; no EMG channels |
| Form factor | Modular forearm capture armband; the video shows the ring structure and inner contact units |
| Sensing array | 10×4 3D force-sensing array |
| Rate | 120 Hz annotated in the video; sampling, transmission refresh and algorithm output should be defined separately, with rates stated per the confirmed hardware and interface version, and distinct from the IMU sampling rate |
| IMU | Inertial sensing and spatial orientation |
| Configuration | Head-mounted stereo + FMG armbands on both forearms |
| Communication | Wireless connectivity and local recording |
| EMG | The current confirmed configuration is FMG + IMU; no EMG sensor is claimed |
The "high-precision / high-sensitivity" claims in the video give no range, accuracy or calibration conditions, so this page does not turn them into quantified performance commitments. 10×4 is not taken to mean 40 or 120 data channels.
| Parameter | Spec / configuration |
|---|---|
| Raw data plan | Raw output of each FMG unit, inertial signals, device timestamps, sequence numbers and quality flags; actual fields subject to interface confirmation |
| Derived outputs | Model predictions such as gestures, 3D hand pose and fingertip pressure; output dimensions, model version, rate and accuracy to be validated |
| Reference ground truth | Hand-pose references, independent force / tactile measurements and human annotation paired per task; reference sensors selected separately |
| Cross-device sync | Clock mapping and sync events saved with Ego / UMI / real-robot data; Thor's in-device sync figures are not reused |
| Wearing records | Anonymous participant ID, left/right side, placement, orientation, tightness, resting baseline, date and re-wearing events |
| Data export plan | Raw data, calibration and derived results saved per session / episode; protocol, SDK license and platform adaptation to be confirmed |
FMG is not EMG; forearm signals are not ground-truth fingertip contact force, and model predictions are not robot actions that can be executed without mapping.
| Parameter | Spec / configuration |
|---|---|
| Sensing performance | Range, sensitivity, accuracy, drift, hysteresis, overload capacity and temperature effects |
| Fit | Forearm circumference range, module count adjustment, preload pressure, long-wear comfort and skin-contact materials |
| Power and structure | Total weight, battery capacity, battery life, charging method, protection rating and cleaning |
| Link performance | End-to-end latency, packet loss, wireless coexistence, reconnection and multi-device concurrency |
| Software and algorithms | Supported systems, raw data access, SDK documentation, model licenses and cross-person calibration requirements |
| Acceptance scope | Fixed tasks, independent test users, cross-day re-wearing, baseline comparisons and safety boundaries |
FROM CAPTURE TO ARCHIVE
Confirm side, module orientation and tightness; complete resting, reference-motion and signal-quality checks as the version requires.
Record task ID, video, FMG, IMU and sync events, keeping original timestamps and never conflating refresh rates of different devices.
Generate supervision from reference poses or independent force measurements, recording model version and uncertainty; raw, reference and predicted data are stored separately.
Compare vision + IMU with and without FMG, splitting train and test sets by user, date and re-wearing, before deciding whether to move on to robot task validation.
BUILD YOUR NEXT SETUP
2 cameras + IMUEGO THOR / 2 CAMERAS + IMU
Multiple configurations
UMI GripperMANIPULATION / TACTILE
Multiple configurations
Tactile Data GloveTACTILE / ELECTROMAGNETIC TRACKING
Multiple configurationsSTART WITH YOUR TASK
From device selection to data and training workflows, we'll help you find the right place to start.