Wrench-ACT: Enhancing Robot Policies for Contact Rich Behavior Using Direct Wrench Control

Johannes Hechtl; Yannik Blei; Simon Ball; Reihaneh Mirjalili; Michael Krawez; Seongjin Bien; Philipp Schmitt; Wolfram Burgard

cs.ROSep 29, 2026

Wrench-ACT: Enhancing Robot Policies for Contact Rich Behavior Using Direct Wrench Control

Authors: Johannes Hechtl, Yannik Blei, Simon Ball, Reihaneh Mirjalili, Michael Krawez, Seongjin Bien, Philipp Schmitt, Wolfram Burgard

Organizations: Siemens Research and Predevelopment · Department of Computer Science and Artificial Intelligence, University of Technology Nuremberg, Germany

Abstract

While contact-rich manipulation requires deliberate regulation of interaction forces, recent approaches to robot manipulation learning predominantly represent actions as target positions or poses. Even methods that incorporate force sensing either use it solely as an observation or, when predicting forces as part of the output, rely on a hybrid force controller. In this paper, we propose an imitation learning policy that predicts wrenches as its sole action output for direct use by a pure force controller. Our studies suggest that force-domain imitation learning depends critically on data collection, with force-feedback teleoperation improving policy performance by capturing the operator's deliberate force regulation. Using Action Chunking with Transformers (ACT) as the base architecture, we train single-task models on bilateral wrench demonstrations and evaluate them on five contact-rich manipulation tasks. The wrench policy matches or outperforms position-based baselines across all tasks, with gains varying according to the degree of deliberate force regulation each task requires. Cross-condition ablations show that the bilateral data collection interface and the wrench action space each contribute independently to performance. To support further research, we will release over 1000 wrench-action demonstrations spanning these tasks on a companion website upon publication.

Figures & tables

Fig. 1 : Method overview of the proposed pure wrench approach. (A) Data collection via bilateral teleoperation. Operator wrench commands are executed by the follower arm and recorded, while the follower position is mirrored back to the leader. (B) Inference. Multimodal observations (RGB streams and robot state) are processed by the policy to directly predict a 6D target wrench, entirely bypassing intermediate positional control for contact-rich manipulation.

Panel of Figure 2 — Fig. 2 : Overview of the five contact-rich manipulation tasks evaluated in this work (a - e) and the ablation evaluating robustness to box-height variation (f)

Fig. 3 : Bilateral Teleoperation Setup. Operator wrench commands, measured by the leader (left), are executed by the follower arm (right). The follower position is mirrored back to the leader.

Fig. 4 : Results of the contact force analysis. Bilateral teleoperation with a wrench action space (

\pi_{\mathrm{b}}^{\mathrm{w}}

) achieves high success rates while maintaining relatively low median 99th-percentile force magnitude and a narrow distribution of applied contact forces.

Action Space		Wrench	Position
Dataset	Bilateral	$\pi_{\mathrm{b}}^{\mathrm{w}}$	$\pi_{\mathrm{b}}^{\mathrm{p}}$
Dataset	VR data	$\pi_{\mathrm{vr}}^{\mathrm{w}}$	$\pi_{\mathrm{vr}}^{\mathrm{p}}$

TABLE I : The four policies resulting from the combination of data collection interface and action representation.

Task	$\pi_{\mathrm{b}}^{\mathrm{w}}$	$\pi_{\mathrm{b}}^{\mathrm{p}}$	$\pi_{\mathrm{vr}}^{\mathrm{w}}$	$\pi_{\mathrm{vr}}^{\mathrm{p}}$
Peg Insertion	74%	2%	2%	6%
Fuse Clipping	88%	72%	52%	96%
Fan Insertion	58%	28%	0%	8%
Industrial Connector	80%	26%	4%	12%
Pen Writing	84%	46%	86%	78%
Average	76.8 %	34.8%	28.8%	40%

TABLE II : Success rates (%) across tasks and policies.

\pi_{\mathrm{b}}^{\mathrm{w}}

: bilateral data with wrench actions;

\pi_{\mathrm{b}}^{\mathrm{p}}

: bilateral data with pose actions;

\pi_{\mathrm{vr}}^{\mathrm{w}}

: VR data with wrench actions;

\pi_{\mathrm{vr}}^{\mathrm{p}}

: VR data with pose actions. All policies are evaluated on 50 rollouts for all tasks.

Fig. 5 : Ablation on the inference frequency on the industrial connector task. We show the dependence of success rate on inference speed. Contrary to a widespread assumption, an inference rate of

50\text{\,}\mathrm{H}\mathrm{z}

is sufficient for a success rate of 80% in our setting.

Task	$\pi_{\mathrm{b}}^{\mathrm{w}}$	$\pi_{\mathrm{b}}^{\mathrm{p}}$	$\pi_{\mathrm{vr}}^{\mathrm{w}}$	$\pi_{\mathrm{vr}}^{\mathrm{p}}$
Pen Writing, Nominal Height	84 %	46%	86%	78%
Pen Writing, Box Raised	76%	8%	56%	24%

TABLE III : Success rates (%) for Pen Writing at the nominal box height and with the box raised by 2.5 cm.

Explore similar work

Learning Dexterous Manipulation Using Contact Wrench Guidance From Human Demonstration
Jun 22, 2026Xinghao Zhu, Zixi Liu, Shalin Jain +17Contact-Rich Dexterous Manipulation Dexterous Manipulation
NVIDIA
Beyond Implicit Force: Evaluating Explicit Force-Torque Proxies in Action Chunking with Transformers
Jul 16, 2026King Hang Wong, Lingqiao Liu, Feras DayoubTorque Teleoperation
Australian Institute for Machine Learning (AIML), Adelaide University, Adelaide, SA 5005, Australia
IMPACT: Learning Internal-Model Predictive Control for Forceful Robotic Manipulation
Jun 9, 2026Jiawei Gao, Chaoqi Liu, Peilin Wu +2Robotic Manipulation Impedance Control
Harvard University · Stanford University

Jun 22, 2026, cs.ROJ/K move · Enter open · S save

Learning Dexterous Manipulation Using Contact Wrench Guidance From Human Demonstration

Xinghao Zhu, Zixi Liu, Shalin Jain, Chenran Li+16

NVIDIA

Contact-Rich Dexterous Manipulation Dexterous Manipulation Robotic Manipulation Contact Wrench Estimation Guidance

Open paper AskPDF arXiv

Dexterous robot manipulation can benefit from the abundance of human demonstrations, but transferring such demonstrations to robot policies remains challenging. We present Contact Wrench Guidance from Human Demonstration in Robotic Dexterous Manipulation (CHORD), a framework for long-horizon manipulation of rigid and articulated objects with reinforcement learning. The key idea is object-centric contact wrench space guidance: we represent human and robot motions by the forces and torques they can induce on the object, enabling similarity to be measured by the induced instantaneous motions. This guidance makes reinforcement learning more scalable for contact-rich dexterous manipulation. We further introduce a large-scale simulation benchmark with 4,739 bimanual dexterous manipulation tasks, constructed from motion-capture datasets and reconstructed in-house videos. Evaluated on 1,831 benchmark tasks, CHORD achieves an average success rate of 82.12%, demonstrating strong scalability. CHORD also generalizes to whole-body manipulation from hand-only and third-person demonstrations, achieving a 90.77% success rate, and the learned policies transfer to the real world in both open-loop and closed-loop settings.

Abstract

Explore similar work

Learning Dexterous Manipulation Using Contact Wrench Guidance From Human Demonstration

Beyond Implicit Force: Evaluating Explicit Force-Torque Proxies in Action Chunking with Transformers

IMPACT: Learning Internal-Model Predictive Control for Forceful Robotic Manipulation