TacSpin: Tactile Tracking of In-Gripper Object Rotation on SO(3) via Axis-Magnitude Estimation

Yuhao Wu; Linhao Song; Huan Yin

cs.ROOct 13, 2026

TacSpin: Tactile Tracking of In-Gripper Object Rotation on SO(3) via Axis-Magnitude Estimation

Authors: Yuhao Wu, Linhao Song, Huan Yin

Organizations: School of AI and Robotics, Hunan University.

Abstract

Tracking object motion within a gripper provides essential feedback for robotic manipulation. However, limited contact coverage and uneven deformation make tactile estimation challenging. In this study, we present TacSpin, a method that tracks in-gripper object rotation on SO(3) using only two vision-based tactile sensors. The key idea is to estimate the axis and magnitude of the rotation response separately, rather than fitting a rigid-body model by least squares. Each finger's response is further refined with contact-boundary cues. The two fingers are then fused according to their agreement. On standard objects, TacSpin achieves lower mean checkpoint errors. Qualitative demonstrations on real-world objects, using calibration transferred from a standard sphere, illustrate tracking under different contact geometries.

Figures & tables

Fig. 1: Object rotation deforms the VBTS elastomer, thereby producing shear and normal responses. TacSpin estimates the axis and magnitude of a rotation response from these observations and accumulates rotation increments on $\mathrm{SO}(3)$ .

Fig. 2: Overview of TacSpin. Each finger provides a rotation response through axis-magnitude estimation and boundary circulation refinement. Agreement-based dual-finger fusion and calibration yield rotation increments that are accumulated on $SO(3)$ .

Fig. 3: Axis-magnitude estimation of a tactile response. Directional voting estimates its axis, and the median of local scalar fits provides its magnitude.

Fig. 4: Boundary circulation refinement of a tactile rotation response. The difference between measured and predicted boundary circulation guides a scaled correction of the initial response.

Fig. 5: Experimental setup. An external camera observes the AprilTag cube attached to the object and the reference tag on the gripper. The inset at the lower right shows the standard objects used for quantitative evaluation and some of the real-world objects used for qualitative demonstrations.

(a) Fixed-axis checkpoint error (5 test sequences per condition)
Condition	TacSE3	TacSE3 + full-K	TacSpin
Cube $Y$ , $30^{\circ}$	$6.5\pm 5.1$	$11.3\pm 7.7$	$\mathbf{5.5}\pm 2.8$
Cube $Y$ , $60^{\circ}$	$17.9\pm 8.1$	$29.1\pm 14.4$	$\mathbf{11.9}\pm 7.7$
Cube $Y$ , $90^{\circ}$	$25.7\pm 8.2$	$43.1\pm 14.8$	$\mathbf{15.3}\pm 10.5$
Cylinder $X$ , $30^{\circ}$	$10.2\pm 4.0$	$8.0\pm 3.8$	$\mathbf{6.8}\pm 3.0$
Cylinder $X$ , $60^{\circ}$	$19.9\pm 3.8$	$17.6\pm 6.6$	$\mathbf{14.1}\pm 9.2$

TABLE I: Rotation errors on standard objects ( ∘ ).

Fig. 6: Rotation tracking errors on the standard objects. (a)–(c) Rotation error for the cube about

Y

and the cylinder about

X

and

Y

, respectively. The horizontal axis is the reference rotation angle relative to the initial orientation, in degrees. Curves show the mean of five test sequences aligned by reference angle over their common angular range; shaded bands indicate

\pm

one sample standard deviation. (d) Mixed-axis sphere motion. Open markers show the mean trajectory error of each test sequence (T06–T08 denote test repetitions 06–08), with gray lines connecting the same sequence across methods. Filled diamonds and error bars show the mean and

\pm

one sample standard deviation across the three sequence-level mean trajectory errors. All plotted sequences are excluded from parameter fitting.

	Mean checkpoint error		Sphere
Variant	Cube	Cylinder	$E_{\mathrm{traj}}$	$e_{R}(t_{\mathrm{end}})$
w/o circ. refinement	17.0	12.5	16.5	30.5
w/o fusion	46.5	55.9	22.4	24.8
Full TacSpin	10.9	11.2	17.3	34.9

TABLE II: Rotation errors of component ablations ( ∘ ).

Fig. 7: Qualitative rotation tracking of different objects. Insets show the corresponding tactile orientation estimates, with the $x$ , $y$ , and $z$ axes in red, green, and blue, respectively.

Tactile Sensing for Robotic Manipulation

Tactile sensing for robotic manipulation uses contact signals to help robots perceive forces, prevent slips, and handle objects reliably.

See all 178 papers

TacSE3: Equivariant SE(3) Motion Estimation from Low-Texture Visuotactile Images for In-Gripper Tracking and Compensation
May 18, 2026Zhongyuan Liao, Junzhe Wang, Qingyang Liu +6Tactile Sensing Object Pose Estimation
Hong Kong Center for Construction Robotics, The Hong Kong University of Science and Technology, Hong Kong, China · The Hong Kong University of Science and Technology (Guangzhou), China · School of Engineering, Great Bay University, Dongguan, China
TacLoc: Global Tactile Localization on Objects from a Registration Perspective
Mar 11, 2026Zirui Zhang, Boyang Zhang, Fumin Zhang +1Object Pose Estimation Point Cloud Registration
ZeroTouch: Tactile-Supervised Visual Contact Estimation for Contact-Rich Manipulation
Sep 18, 2026Dmitriy Kosenkov, Daniia Zinniatullina, Miguel Altamirano Cabrera +3Robotic Grasping Contact-Rich Robotic Manipulation
Intelligent Space Robotics Laboratory, Skolkovo Institute of Science and Technology, Moscow, Russian Federation.

May 18, 2026J/K move · Enter open · S save

TacSE3: Equivariant SE(3) Motion Estimation from Low-Texture Visuotactile Images for In-Gripper Tracking and Compensation

Zhongyuan Liao, Junzhe Wang, Qingyang Liu, Zhenmin Huang+5

Hong Kong Center for Construction Robotics, The Hong Kong University of Science and Technology, Hong Kong, China · The Hong Kong University of Science and Technology (Guangzhou), China · School of Engineering, Great Bay University, Dongguan, China

Tactile Sensing Object Pose Estimation Robotic Manipulation Tactile Sensing for Robotic Manipulation

Open paper AskPDF arXiv

Robotic in-hand manipulation requires reliable object-motion tracking under frequent visual occlusion, yet low-texture visuotactile images provide few stable correspondences for conventional image- or geometry-matching methods. This paper presents TacSE3, a tactile motion-estimation pipeline that converts low-texture visuotactile observations into a decoupled three-dimensional force field and estimates incremental rigid-body motion on SE(3). The method derives planar translation from contact-centroid motion and estimates rotation primarily from shear-related tactile responses, yielding a physically interpretable signal for in-gripper tracking and compensation. Experiments with paired DM-Tac fingertip sensors show that dual-sensor sensing reduces translation-rotation ambiguity, supports rotation tracking across axes and object geometries, and provides a lightweight compensation signal that improves disturbance tolerance in downstream manipulation tasks without retraining the base policy.

Abstract

Tactile Sensing for Robotic Manipulation

Explore similar work

TacSE3: Equivariant SE(3) Motion Estimation from Low-Texture Visuotactile Images for In-Gripper Tracking and Compensation

TacLoc: Global Tactile Localization on Objects from a Registration Perspective

ZeroTouch: Tactile-Supervised Visual Contact Estimation for Contact-Rich Manipulation