Proprioceptive Finger Force Estimation for Reliable Box Manipulation

Junha Cha; Chanwoo Kim; Kisang Park; Yeoncheol Jang; Doyoung Kim; Seungmin Kim; Kyungjae Lee; Sungjoon Choi

cs.ROOct 28, 2026

Proprioceptive Finger Force Estimation for Reliable Box Manipulation

Authors: Junha Cha, Chanwoo Kim, Kisang Park, Yeoncheol Jang, Doyoung Kim, Seungmin Kim, Kyungjae Lee, Sungjoon Choi

Organizations: Department of Artificial Intelligence, Korea University, Seoul, Republic of Korea. · Department of Statistics, Korea University, Seoul, Republic of Korea. · RLWRLD, Seoul, Republic of Korea.

Abstract

Reliable box manipulation requires successful placement without damage, which depends on maintaining sufficient contact force without excessive loading. We hypothesize that per-finger force feedback improves manipulation reliability relative to open-loop finger commands. To this end, we present Finger External-Force Estimation from Proprioceptive SignaLs (FEEL), an uncertainty-aware residual method that estimates per-finger external force magnitude from proprioceptive histories. Our method combines a calibrated virtual-spring estimate with a residual distribution predicted by a Mixture Density Network (MDN) and requires no dedicated contact sensors during deployment. The estimated force is used for continuous per-finger adaptation, while predictive uncertainty is evaluated as an empirical signal of ambiguity in proprioceptive force estimation. Across ten fingers and contact forces spanning 0-20 N, our method achieves a mean RMSE of 0.251 N, reducing the error by 92.8% relative to the virtual spring and by 9.4% relative to an end-to-end MDN. In box pick-and-place, force adaptation maintains task success at 10/10 and increases both damage-free outcomes and reliability from 7/10 to 10/10. The project page is available at https://finger-force-estimation.pages.dev/

Figures & tables

Fig. 1 : Overview of FEEL. (A) A wearable F/T sensor collects reference force magnitudes synchronized with proprioceptive histories during finger motion. (B) The proprioceptive history $\mathbf{x}_{i,t}$ serves as input to FEEL. The calibrated virtual spring provides the model-based force estimate $f^{\mathrm{spr.}}_{i,t}$ , while the MDN predicts the residual $r_{\theta}(\mathbf{x}_{i,t})$ and its predictive uncertainty. (C) The two force terms are added to obtain the per-finger estimate, $\hat{f}_{i,t}=f^{\mathrm{spr.}}_{i,t}+r_{\theta}(\mathbf{x}_{i,t})$ . During deployment, this estimate provides feedback for force-based finger adaptation without dedicated contact sensors.

Fig. 2 : Real force-data collection using a wearable F/T sensor. Contact interactions at multiple finger locations are synchronized with proprioceptive measurements.

Fig. 3 : Force distributions in the collected dataset for ten fingers. The top and bottom rows show the distributions of force data collected from the left and right hands, respectively. Within each finger, wider regions indicate more samples at that force level. White circles mark the median force, and black bars span the middle 50% of samples.

Hand	Finger	Virtual spring [ 13 ]	End-to-end MDN [ 22 ]	Residual MDN (Ours)
Hand	Finger	RMSE (N) $\downarrow$
Left	Thumb	$5.327\pm 0.012$	$0.790\pm 0.029$	0.723 $\pm$ 0.024
	Index	$3.203\pm 0.042$	$0.192\pm 0.030$	0.178 $\pm$ 0.044
	Middle	$2.545\pm 0.026$	0.093 $\pm$ 0.002	$0.096\pm 0.004$
	Ring	$3.467\pm 0.030$	$0.140\pm 0.059$	0.119 $\pm$ 0.021
	Little	$2.135\pm 0.033$	0.103 $\pm$ 0.004	$0.106\pm 0.006$

TABLE I : Comparison of per-finger external force magnitude estimation accuracy using row-level five-fold cross-validation stratified by force magnitude bins.

Fig. 4 : Comparison of force magnitude traces from matched tapping motions in MuJoCo and on the real robot.

Hand	Residual MLP	Residual MDN (Ours)
Hand	RMSE (N) $\downarrow$
Left	$0.261\pm 0.004$	0.245 $\pm$ 0.008
Right	$0.280\pm 0.004$	0.257 $\pm$ 0.016
Overall	$0.270\pm 0.004$	0.251 $\pm$ 0.010

TABLE II : Ablation of residual output models for external force magnitude estimation.

Fig. 5 : Snapshots of real-robot box pick-and-place with and without force-based finger adaptation. The enlarged views show box deformation and damage at the designated seam without adaptation, whereas the seam remains intact with adaptation.

Method	Success $\uparrow$	Damage- free $\uparrow$	Reliability $\uparrow$
No force adaptation	10/10	$7/10$	$7/10$
Force adaptation (Ours)	10/10	10/10	10/10

TABLE III : Comparison of force adaptation under one predefined box pick-and-place motion over 10 trials per condition.

	RMSE (N) $\downarrow$	Box manipulation (/10)
$(K_{p},K_{d})$	RMSE (N) $\downarrow$	Success $\uparrow$	Damage- free $\uparrow$	Reliability $\uparrow$
$(100,200)$	0.233 $\pm$ 0.005	4/10	10/10	4/10
$(200,400)$	$0.251\pm 0.010$	10/10	10/10	10/10
$(300,600)$	$0.255\pm 0.008$	9/10	8/10	7/10

TABLE IV : Controller-gain sensitivity of force magnitude estimation and force-adaptive box manipulation.

Interaction condition	Aleatoric uncertainty	Epistemic uncertainty
Force through joints	0.370 $\pm$ 0.001	0.102 $\pm$ 0.005
Force through structure	0.364 $\pm$ 0.005	0.172 $\pm$ 0.088

TABLE V : Comparison of predictive uncertainty for Force through joints and Force through structure.

Contact Force Estimation

See all 28 papers

Multifingered force-aware control for humanoid robots
Mar 9, 2026Pasquale Marra, Gabriele M. Caddeo, Ugo Pattacini +1Humanoid Robot Control Force-Controlled Robotic Manipulation
Humanoid Sensing and Perception, Istituto Italiano di Tecnologia, Genoa, Italy. · DIBRIS, Universit`a di Genova, Via All’Opera Pia, 13, Genoa, Italy. · MESH Facility, Istituto Italiano di Tecnologia, Genoa, Italy
Current as Touch: Proprioceptive Contact Feedback for Compliant Dexterous Manipulation
Jul 3, 2026Chenyang Ma, Yunchao Yao, Zhenyu Wei +3Robotic Control Compliant Robot Control
University of North Carolina at Chapel Hill
Force/Torque-Based Kinematic Adaptation for Robotic Manipulation Tasks
Aug 21, 2026Carl Glen HenshawRobotics Robotic Control
US Naval Research Laboratory · U.S. Naval Research Laboratory

Mar 9, 2026J/K move · Enter open · S save

Multifingered force-aware control for humanoid robots

Pasquale Marra, Gabriele M. Caddeo, Ugo Pattacini, Lorenzo Natale

Humanoid Sensing and Perception, Istituto Italiano di Tecnologia, Genoa, Italy. · DIBRIS, Universit`a di Genova, Via All’Opera Pia, 13, Genoa, Italy. · MESH Facility, Istituto Italiano di Tecnologia, Genoa, Italy

Humanoid Robot Control Force-Controlled Robotic Manipulation Robotic Manipulation Contact Force Estimation Tactile Sensing for Robotic Manipulation

Open paper AskPDF arXiv

In this paper, we address force-aware control and force distribution in robotic platforms with multi-fingered hands. Given a target goal and force estimates from tactile sensors, we design a controller that adapts the motion of the torso, arm, wrist, and fingers, redistributing forces to maintain stable contact with objects of varying mass distribution or unstable contacts. To estimate forces, we collect a dataset of tactile signals and ground-truth force measurements using five Xela magnetic sensors interacting with indenters, and train force estimators. We then introduce a model-based control scheme that minimizes the distance between the Center of Pressure (CoP) and the centroid of the fingertips contact polygon. Since our method relies on estimated forces rather than raw tactile signals, it has the potential to be applied to any sensor capable of force estimation. We validate our framework on a balancing task with five objects, achieving a

82.7\%

success rate, and further evaluate it in multi-object scenarios, achieving

80\%

accuracy. Code and data can be found here https://github.com/hsp-iit/multifingered-force-aware-control.

Abstract

Contact Force Estimation

Explore similar work

Multifingered force-aware control for humanoid robots

Current as Touch: Proprioceptive Contact Feedback for Compliant Dexterous Manipulation

Force/Torque-Based Kinematic Adaptation for Robotic Manipulation Tasks