CoinFT: A Coin-Sized, Capacitive 6-Axis Force Torque Sensor for Robotic Applications

Hojung Choi; Jun En Low; Tae Myung Huh; Seongheon Hong; Gabriela A. Uribe; Kenneth A. W. Hoffmann; Julia Di; Tony G. Chen

cs.ROMar 25, 2025

CoinFT: A Coin-Sized, Capacitive 6-Axis Force Torque Sensor for Robotic Applications

Authors: Hojung Choi, Jun En Low, Tae Myung Huh, Seongheon Hong, Gabriela A. Uribe, Kenneth A. W. Hoffmann, Julia Di, Tony G. Chen, +2 more

Organizations: Stanford University, CA, USA. · University of California Santa Cruz, CA, USA. · Reality Labs Research, Meta Platforms Inc, WA, USA.

Abstract

We introduce CoinFT, a capacitive 6-axis force/torque (F/T) sensor that is compact, light, low-cost, and robust with an average root-mean-squared error of 0.16 N for force and 1.08 mN m for moment when the input ranges from 0-14 N and 0-5 N in normal and shear directions, respectively. CoinFT is a stack of two rigid PCBs with comb-shaped electrodes connected by an array of silicone rubber pillars. A microcontroller interrogates the electrodes in different subsets in order to enhance sensitivity for measuring 6-axis F/T. The combination of features of CoinFT enables various contact-rich robot interactions across different embodiment domains including drones, robot end-effectors, and wearable haptic devices. We demonstrate the utility of CoinFT through two representative applications: a multi-axial contact-probing experiment in which a CoinFT mounted beneath a hemispherical fingertip measures 6-axis force and torque representative of manipulation scenarios, and a force-control task on a drone. The design, fabrication, and firmware of CoinFT are open-sourced at https://coin-ft.github.io/.

Figures & tables

Sensor	Transduction	Size [mm]	Mass	Sensing range	Accuracy ( $R^{2}$ unless noted otherwise)	Bandwidth	Cost	Robustness
MinebeaMitsumi™ 6-axis F/T sensor	Piezoresistive	$\phi$ 9.6 $\times$ $t$ 9	3 g	$F$ : 40 N $M$ : 400 mN·m	Full-scale error: $\pm$ 5%	1 kHz	$1200+	No
Guggenheim et al. [ 10 ]	Barometric (MEMS)	$75\times 75\times 30$	N/A	$F_{x,y}$ : 5 N $F_{z}$ : 12 N $M$ : 3000 mN·m	$F_{x}$ : 0.937 $F_{y}$ : 0.909 $F_{z}$ : 0.998 $M_{x}$ : 0.984 $M_{y}$ : 0.984 $M_{z}$ : 0.991	50 Hz	$20	Yes
Ouyang et al. [ 15 ]	Vision (Fiducial tracking)	$36\times 31\times 51$	N/A	$F_{x,y}$ : 2 N $F_{z}$ : 1 N $M_{x,y}$ : 100 mN·m $M_{z}$ : 4 mN·m	$F_{x}$ : 0.991 $F_{y}$ : 0.996 $F_{z}$ : 0.875 $M_{x}$ : 0.997 $M_{y}$ : 0.998 $M_{z}$ : 0.902	25 Hz (sampling)	$40	Yes
Kim et al. [ 16 ]	Capacitive	$14\times 17\times 6.5$	2.6 g	$F$ : 30 N $M$ : 300 mN·m	Full-scale error: $Fx$ : 0.92% $Fy$ : 0.83% $Fz$ : 1.57% $Mx$ : 2.8% $My$ : 2.1% $Mz$ : 3.4%	200 Hz (sampling) $<$ 5 Hz (mech. estimated)	N/A	N/A
El-Azizi et al. [ 17 ]	Optical	$\phi$ 27 $\times$ $t$ 20	16.4 g	$F$ : 10 N $M$ : 200 mN·m	$F_{x}$ : 0.990 $F_{y}$ : 0.993 $F_{z}$ : 0.560 $M_{x}$ : 0.995 $M_{y}$ : 0.991 $M_{z}$ : 0.923	2.5 kHz (sampling)	$100	Yes
Do et al. [ 18 ]	Vision (Raw image)	32 $\times$ 32 $\times$ 43	34 g	$F_{x,y}$ : 5.5 N $F_{z}$ : 11 N $M_{x,y}$ : 250 mN·m $M_{z}$ : 50 mN·m	Mean Avg. Error: $F$ : 0.41 N $M$ : 0.387 mN·m	30 Hz (sampling)	$78	N/A

TABLE I: Comparison of representative lower-cost 6-axis force/torque sensors for robotic applications

Fig. 1: (a) CoinFT is approximately the size of a U.S. quarter-dollar coin. (b) Exploded view showing two rigid PCBs (upper and lower sensing layers , with eight and two individually addressable sensing electrodes, respectively ) connected with an array of silicone rubber pillars. The fPCB top shield layer provides passive shielding. Alignment tabs are removed after assembly. (c) Layer stackup and dimensions of each layer. The overall thickness is $\approx 2$ mm.

Fig. 2: (a) The microcontroller unit (MCU) firmware switches between two electrode configurations (Normal Mode and Shear Mode) using the analog switches, which connect each electrode to the capacitance-to-digital conversion (CDC) module, to ground (GND), or to an active shield. In the instant depicted in the schematic, Z1 is being read in Normal Mode while X1 is being read in Shear Mode. The firmware cycles through all electrodes in both modes to complete one full sampling cycle. (b) The first three columns (no input, shear force, moment) show the resulting capacitance patterns in plan view; the last two (normal force, moment) show them in cross-section. Ground and Shield are the two electrodes of the lower sensing layer (see legend). These configurations produce distinct signal patterns under different force and torque inputs.

Fig. 3: The fabrication process of CoinFT. (a) Fresh uncured silicone is spread on a UV laser cut mask that is placed on an acrylic plate with alignment pins. (b) A primed lower sensing layer PCB and an acrylic plate for equal pressure distribution is stacked with a 3.1 kg weight. (c) With a 0.1 kg weight, the assembly is cured inside a pressurized chamber. (d),(e) Once the mask and acrylic plates are removed, the pillar layer is complete. (f) On a primed upper sensing layer PCB, fresh uncured silicone is spread. (g) The silicone layer is made thin and uniform through spin-coating. (h) The lower sensing layer with pillars is assembled with the upper sensing layer through precise distance control by adding spacers. A 0.1 kg weight and an acrylic plate for pressure distribution are added. (i) The top shield layer is attached using an adhesive. (j) A horizontal cross section of the pillar layer shows desirable bonding of the two PCB layers.

Item	Quantity	Unit Cost [$]	Total Cost [$]
PSoC 4100S	1	3.25	3.25
Upper sensing Layer PCB	1	2.94	2.94
Lower sensing layer PCB	1	0.76	0.76
Shield PCB	1	2.09	2.09
SMD electronics	6	0.20	1.20
Molex connector	1	0.69	0.69

TABLE II: Bill of Materials (BOM) for CoinFT

Fig. 4: CoinFT characterization using FEA. (a) Structural modeling in FEA. (b) Electrostatics modeling in FEA. (c) Raw capacitance change with displacement and corresponding compressive force. Full CoinFT response with varied pillar diameter under (d) normal force (real samples & FEA), (e) shear force (FEA), and (f) torsion (FEA). Raw capacitance response of a half of CoinFT (g) under normal force, (h) shear force, and (i) torsion.

Fig. 5: Comparison of sensor readings between CoinFT and Gamma (ATI Industrial Automation) in (a) $F_{x}$ , (b) $F_{y}$ , (c) $F_{z}$ , (d) $M_{x}$ , (e) $M_{y}$ , (f) $M_{z}$ .

Input Range: 0 $\sim$ 5 N Normal, 0 $\sim$ 2 N Shear
	$F_{x}$	$F_{y}$	$F_{z}$	$M_{x}$	$M_{y}$	$M_{z}$
Normal+Shear (RMSE)	0.038	0.040	0.050	0.317	0.306	0.231
Shear (RMSE)	0.039	0.040	0.115	0.343	0.323	0.244
Normal+Shear ( $R^{2}$ )	0.994	0.994	0.997	0.995	0.995	0.992
Shear ( $R^{2}$ )	0.993	0.994	0.986	0.994	0.994	0.992
Input Range: 0 $\sim$ 14 N Normal, 0 $\sim$ 5 N Shear

TABLE III: Accuracy of CoinFT calibration.

Fig. 7: Effects of temperature variation and its compensation. (a) Drift of raw signals due to temperature change. (b) Temperature compensation using an on-chip temperature sensor.

Fig. 8: CoinFT robustness evaluation. (a) CoinFT continues to provide reliable force readings even after an impact from a hammer. (b) The calibration stays consistent after ten impacts from a 0.53 kg mass dropped at a height of 0.63 m.

	$F_{x}$	$F_{y}$	$F_{z}$	$M_{x}$	$M_{y}$	$M_{z}$
RMSE	0.108	0.120	0.378	5.669	3.874	0.720
$R^{2}$	0.994	0.995	0.999	0.993	0.996	0.986

TABLE IV: Multi-axial sensing during contact probing.

Fig. 9: Multi-axial contact probing experiment. (a) The 7DoF arm applies force and torque in different axes on a hemispherical fingertip equipped with CoinFT, approximating manipulation scenarios. (b) CoinFT and reference sensor reading comparisons. Refer to Table IV for quantitative evaluation.

Fig. 10: Experimental setup for drones performing contact rich tasks. Different end-effectors can be attached to the CoinFT mount, such as (a) a tip for general contact or (b) a package of electronics to be attached on a surface. Compliance is added in series through a (c) telescoping mechanism with pre-tensioned rubber bands to mitigate impact upon contact.

Fig. 13: Drone attaching a package of electronics on environment surfaces using force control. (a) The drone gently presses the package on the horizontal surface. (b) After descending, it still feels the weight of the package through CoinFT. (c) The drone presses the package with a larger force on the surface. (d) Upon descent, it no longer feels the weight of the package. It turns on the package and leaves the scene.

Explore similar work

Jul 10, 2026cs.RO

PinFT: Miniature 5-Axis Force/Torque Sensor Embeddable to Tweezer-like Tool

We present PinFT, a miniature five-axis capacitive force/torque sensor designed for direct tip-level integration into tweezer-like tools. The sensor employs a compact three-PCB stack with segmented plated through-hole electrodes and a silicone elastomer dielectric, enabling five-degree-of-freedom force and torque sensing (

F_x

F_y

F_z

T_x

T_y

) through displacement of a central 2,mm-diameter stainless steel pin. The fabricated prototype was calibrated using a higher-order polynomial mapping, yielding mean absolute errors of approximately 0.23,N for forces and 2.5,mN

\cdot

m for torques, with coefficients of determination (

R^2

) exceeding 0.97 across all axes. To demonstrate practical utility, a 3D-printed tweezer integrating PinFT sensors at both tips was mounted on a parallel-jaw gripper and evaluated across three representative manipulation tasks: grasping a sub-millimeter SMD capacitor, pulling a simulated hair from a silicone substrate, and tearing a compliant silicone specimen. In all cases, per-tip force sensing reliably captured characteristic force signatures that distinguish successful manipulation from failure events -- including slip and object ejection -- using gradient-based features derived from internal grasp force and net interaction force. These results demonstrate that direct, per-tip force sensing enables standard parallel-jaw grippers to monitor and interpret fine manipulation tasks performed through a handheld tweezer.

Leo King, Jenny Chen, Tae Myung Huh

Dept. of Electrical and Computer Engineering, University of California Santa Cruz, Santa Cruz, CA, USA.

Ground Reaction Forces Gripper Force Torque Morphogenesis

Ask PDF

Jul 17, 2026cs.RO

Scalable Open-Source Visuotactile Sensor for 6-Axis Contact Wrench Estimation in Tensegrity Robots

This paper presents a scalable, open-source visuotactile sensing system for tensegrity robots that enables six-axis wrench estimation and contact detection. The proposed endcap sensor integrates an elastomeric shell, a 3D-printed thermoplastic polyurethane (TPU) interface, and a rigid base housing an embedded camera and LED illumination ring. A novel gyroid-infill bonding technique is introduced to form a durable elastomer-TPU interface without adhesives, yielding a lightweight and modular design compatible with large-scale tensegrity structures. A tactile-to-wrench neural network maps shear vector fields to six-dimensional force and torque measurements. Experimental results demonstrate accurate and stable wrench estimation with a mean squared error (MSE) of 0.1531 on static validation data and out-of-domain generalization under dynamic motion. Furthermore, full-system integration on a 12 kg tensegrity robot confirms the sensor's ability to reliably identify ground contacts. The system substantially improves the practicality of tactile feedback for tensegrity robots, offering a low-cost, reproducible, and physically interpretable pathway toward contact-aware proprioception and state estimation. Open source files are available at \href{https://github.com/Jonathan-Twz/tensegrity-gelfoot}{github.com/Jonathan-Twz/tensegrity-gelfoot}

Wenzhe Tong, Jonathan Mi, Xili Yi +2

Robotics Department, University of Michigan, Ann Arbor, MI, 48109, USA

Tactile Contact Wrench Estimation Proprioception Sensors

Ask PDF

Jun 10, 2026cs.RO

FACTR 2: Learning External Force Sensing for Commodity Robot Arms Improves Policy Learning

Contact-rich manipulation requires force sensitivity, but many robot arms lack dedicated force sensors due to their high cost. We present Neural External Torque Estimation (NEXT), a data-driven method that estimates external joint torques without needing any dedicated force sensors. NEXT trains in 1 minute from only 10 minutes of free-motion data, yet achieves estimates comparable to dedicated joint-torque sensors. NEXT enables force-feedback teleoperation on low-cost arms and improves policy learning through Force-Informed Re-Sampling Training (FIRST), which up-samples pre-contact and contact segments during behavior cloning. Across five long-horizon tasks, FIRST outperforms prior force-aware policies by over 17% in task progress. Together, NEXT and FIRST bring force-aware teleoperation and policy learning to off-the-shelf robots without additional sensing hardware. Video results and code are available at https://jasonjzliu.com/factr2

Steven Oh, Jason Jingzhou Liu, Tony Tao +5

Carnegie Mellon University · Waseda University

Ground Reaction Forces Torque Teleoperation Soft Robotics Robot Systems Factor

Ask PDF