Introduction
Vision can locate an object before contact, but touch is what reveals local pressure, shear, slip and load redistribution after the fingers close. Tactile hands therefore differ not only by sensor count but by coverage, calibration, bandwidth and how quickly feedback changes the grasp.
Current 1X, Unitree and Figure hands publish different tactile specifications and test methods. This page keeps those manufacturer claims separate instead of converting them into a single tactile-performance ranking.
Key findings
- 1X's July 2026 NEO hand release specifies 25 DoF in total, including 22 fully actuated finger-and-palm DoF plus three wrist axes, with tactile and shear sensing across the fingers.
- Unitree's Dex3-1 official specification lists seven DoF, 33 pressure sensors, a 10 g to 2,500 g sensing range under its stated test method and 1,000 Hz communication.
- Figure says Figure 03's fingertip sensors can detect forces associated with a three-gram object; this remains a company specification rather than a cross-vendor benchmark.
- Tactile performance depends on coverage, calibration, bandwidth, contact geometry and how quickly the controller reacts to slip or load redistribution.
- Published tactile specifications are not directly comparable across different skin technologies and test methods.
Robot Hands With Tactile Sensors What They Can Measure - evidence comparison
The table records what each source establishes and keeps missing data visible.
| System or method | What the evidence establishes | Evidence class | Main unresolved point |
|---|---|---|---|
| 1X NEO | The company describes tactile and shear sensing across a tendon-driven hand intended for home manipulation. | Company documentation | Sensor area, sampling rate and calibration methods are often unpublished. |
| Unitree Dex3-1 | The official page lists 33 pressure sensors on a three-finger hand. | Officially documented | Company demonstrations rarely include raw tactile traces. |
| Figure 03 | Figure states its first-generation tactile system can detect forces associated with a three-gram object; the claim is not an independent benchmark. | Company claim | Tactile performance is not directly comparable across skins and contact geometries. |
| Shadow Dexterous Hand | Research configurations support tactile fingertips and force sensing with detailed technical documentation. | Commercial research evidence | Sensor area, sampling rate and calibration methods are often unpublished. |
Evidence classification
- Officially documented specifications, standards or project status stated by the responsible organization.
- Real-system evidence demonstrations or deployments performed on physical hardware under described conditions.
- Company claim a numerical or operational statement reported by the company and not independently audited.
- Simulation or research evidence useful for mechanisms, but not proof of field deployment.
- Insufficient public evidence control mode, trial count, version or operating conditions are missing.
Definition and design boundary
Tactile sensing converts physical contact at the skin, fingertip, palm or wrist into signals used by estimation and control. It includes pressure arrays, capacitive cells, piezoresistive elements, optical tactile images, force-torque sensors and motor-current estimates. A force sensor at the wrist is not equivalent to distributed fingertip touch.
How the hand architecture works
Calibrate raw sensor values against force, location and temperature. Fuse tactile data with vision, joint position and motor current. Detect contact onset, slip direction and load redistribution. Use fast local control loops to adjust grip before the object falls. Record synchronized tactile-action sequences for policy training. Latency, calibration and safety limits can change the result even when the high-level model remains the same.
How to compare dexterity claims
The evidence check for robot hand with tactile sensors starts with 1X Technologies, Unitree Robotics, Figure AI and works outward only when a primary record is incomplete.
Failure modes during manipulation
Failure modes include sensor drift, skin wear, temperature sensitivity, sparse coverage that misses edge contact or torsion, and latency between contact change and grip correction. Soft skins can spread force and complicate localization, while optical tactile systems add camera and illumination dependencies.
Credible applications today
Credible applications include Grip stabilization, slip recovery and insertion, Fragile-object handling and force-limited grasping, Contact-rich assembly when the contact point is visually occluded and Tactile-language-action research that conditions a policy on touch and instructions. These applications should be described with the robot, task boundary, operator role and environmental constraints. Experimental capability, commercial availability and routine deployment are reported as separate statuses.
Questions buyers and researchers should ask
Ask which physical sensor is installed, where it covers the hand, how it is calibrated and how quickly a contact change reaches the controller. Then separate the sensor from the learned model a reusable tactile encoder, a contact predictor and a tactile-language-action policy produce different outputs and require different evidence.
Tactile foundation models are a distinct 2026 research layer above the hardware described here. Work such as Sparsh, AnyTouch and TouchWorld studies transferable touch representations or predictive contact dynamics. Those models may improve slip and alignment correction, but they do not make sensor drift, skin wear, force limits or hand-specific calibration disappear.
Limitations and missing information
- Sensor area, sampling rate and calibration methods are often unpublished.
- Company demonstrations rarely include raw tactile traces.
- Tactile performance is not directly comparable across skins and contact geometries.
- Specifications, prices, repositories and deployment status can change after publication.
- Benchmarks from different robots or environments are not directly comparable.
Conclusion
Tactile sensing is most useful when it closes a fast control loop for contact, slip and load redistribution. Current hands publish very different sensing architectures and test methods, so sensor count alone is not a dexterity score. Buyers and researchers should compare coverage, calibration, bandwidth, raw feedback access and repeated task performance on the same contact problem.
Frequently asked questions
How should a tactile robot hand be evaluated?
Record sensor location and coverage, calibrated force or pressure range, sampling and communication rate, drift, temperature sensitivity, raw feedback access and how quickly contact changes alter the grip. Then test repeated slip, insertion and fragile-object tasks rather than comparing sensor counts alone.
Sources and methodology
Sources for tactile robot hands were rechecked on September 27, 2026 using current official material from 1X, Unitree and Figure. 1X's 25-DoF hand, Unitree Dex3-1's seven-DoF/33-sensor specification and Figure 03's three-gram sensitivity claim remain attributed to their manufacturers and are not treated as one common benchmark.
Related TechniaHQRobot guides
Source review notes
Review date recorded July 11, 2026
Supported by the cited sources
- 1X NEO: The company describes tactile and shear sensing across a tendon-driven hand intended for home manipulation.
- Unitree Dex3-1: The official page lists 33 pressure sensors on a three-finger hand.
Not confirmed or incomplete
- Sensor area, sampling rate and calibration methods are often unpublished.
- Company demonstrations rarely include raw tactile traces.
- Tactile performance is not directly comparable across skins and contact geometries.
Likely to change quickly
- Commercial availability, prices, model versions and software access.
- Deployment counts, company partnerships and repository maintenance status.
Share this article
Share the current TechniaHQRobot article page.
Continue reading
Open the latest robotics reporting, Physical AI analysis and hardware notes.