TacForcing: Streaming Action Generation with Execution-Time Tactile Feedback

Jianbo Zhou; Boyuan Zhao; Yuzheng Zhang; Yiyang Chen; Wenxin Chen; Qiuyue Li; Xiangyang Gu; Yuhan Cao

cs.ROAug 26, 2026

TacForcing: Streaming Action Generation with Execution-Time Tactile Feedback

Authors: Jianbo Zhou, Boyuan Zhao, Yuzheng Zhang, Yiyang Chen, Wenxin Chen, Qiuyue Li, Xiangyang Gu, Yuhan Cao, +4 more

Organizations: School of Computer Science, SJTU · SCUT · ECUST · ECNU · PolyU · CCUT · UESTC · HUST

Abstract

Contact-rich manipulation requires adapting to contact states that can evolve substantially within an action horizon. However, chunk-based vision-language-action models predict complete action chunks from observations collected before execution, leaving tactile conditioning stale during execution. Existing tactile-reactive approaches typically rely on separate high-frequency controllers, which increase both architectural and training complexity. In this paper, we introduce TacForcing, a streaming action-generation framework incorporating execution-time tactile feedback. TacForcing replaces the standard action expert with a streaming expert that generates action blocks sequentially while preserving intermediate states of unfinished blocks. After each block is executed, the expert resumes generation from these states using newly acquired tactile feedback. To better align tactile conditioning with action execution, we further introduce Execution-Aware Tactile Attention (EATA), which restricts direct access to each tactile update to the next block scheduled for execution. Across six UniVTAC simulation tasks and six real-world contact-rich manipulation tasks on two robot platforms, TacForcing achieves average success rates of 65% in simulation and 66% in the real world, outperforming the strongest baselines by 6 and 15 percentage points, respectively.

Figures & tables

Figure 2 : Visual and tactile dynamics during dropper squeezing. (a) Visual observations at the beginning and end of the action horizon. (b) Deformation maps from the thumb and index finger, sampled every five actions. (c) Cosine distances from the initial visual and tactile representations; the inset provides an enlarged view of the visual cosine distances.

Figure 3 : Overview of TacForcing . The VLM encodes the visual observation and language instruction once, and the resulting task context is reused throughout streaming generation. The Streaming Action Expert progressively refines action blocks according to block-specific flow times, allowing successive blocks to become ready for sequential execution. After each ready block is executed, the resulting tactile feedback is encoded before generation resumes from the retained intermediate states of unfinished blocks. Execution-Aware Tactile Attention restricts direct access to the latest tactile feedback to the block scheduled for execution next, thereby reducing the temporal mismatch between tactile acquisition and action execution.


Method	Lift Bottle	Pull-out Key	Lift Can	Put Bottle in Shelf	Insert Hole	Insert Tube	Avg.

$\pi_{0.5}$	88	43	46	43	39	48	51
UniVTAC-ACT	59	41	24	6	36	58	37
RDP	84	18	12	41	23	75	42
FTP-1	89	35	66	23	62	76	59

Table 1 : Results on the UniVTAC simulation benchmark. Success rates (%). The highest and second-highest distinct values in each column are shown in bold and underlined , respectively.

Figure 4 : Real-world platforms and tasks. Top: Close Cap , Insert Battery , and Close Lid on Sharpa North. Bottom: Transfer Liquid , Wipe Board , and Stand Bottle on Sharpa & RealMan. Platform views are shown on the left; task scenes are shown on the right.

Method	Close Cap	Insert Battery	Close Lid	Transfer Liquid	Wipe Board	Stand Bottle	Avg.
	Sharpa North			Sharpa & RealMan
$\pi_{0.5}$	25	38	13	6	31	44	26
GR00T N1.7	44	75	25	19	50	56	45
FTP-1	50	63	38	6	75	75	51
T-Rex	44	50	44	31	63	75	51
TacForcing (Ours)	56	81	50	50	75	81	66

Table 2 : Results on six real-world tasks. All values are success rates (%). The highest and second-highest distinct rates in each column are shown in bold and underlined , respectively.

Figure 5 : Ablation results. All variants retain streaming generation. Fixed tactile holds the initial tactile observation constant within each action chunk, while tactile updates refresh the observation during execution; both disable EATA. TacForcing enables both updates and EATA. Average denotes the mean success rate over the three tasks in each panel.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6 : Real-world platform. The setup comprises (a) Sharpa Wave dexterous hands, (b) Manus Pro data gloves, (c) two RealMan RM75 robot arms, and (d) top- and wrist-mounted RealSense cameras.

Figure 7 : Sharpa North tasks. Representative execution sequences in the task order of Table 2 . Each row progresses from left to right.

Figure 8 : Sharpa & RealMan tasks. Representative execution sequences in the task order of Table 2 . Each row progresses from left to right.

Figure 9 : UniVTAC simulation tasks. Representative execution sequences for the six tasks used in our simulation experiments, reproduced from UniVTAC [ 4 ] .

Explore similar work

Dream-Tac: A Unified Tactile World Action Model for Contact-Rich Robot Manipulation
Jun 7, 2026Yunfan Lou, Yifan Ye, Yankai Fu +7Tactile World Model Contact-Rich Dexterous Manipulation
Peking University Beijing, China · Nanjing University Nanjing, China · State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University Beijing, China
TacForeSight: Force-Guided Tactile World Model for Contact-Rich Manipulation
Jun 9, 2026Yujie Zang, Yuhang Zheng, Xian Nie +7Tactile World Model Tactile
TARS Robotics · National University of Singapore · Shanghai Jiao Tong University +2
ReTouch: Empowering Contact-Rich Dexterous Manipulation with Online-Refined Tactile Prediction
Aug 3, 2026Shiqi Zhang, Xin Zhang, Yedong Shen +9Contact-Rich Dexterous Manipulation Tactile
University of Science and Technology of China · The Chinese University of Hong Kong · iFLYTEK

Jun 7, 2026, cs.ROJ/K move · Enter open · S save

Dream-Tac: A Unified Tactile World Action Model for Contact-Rich Robot Manipulation

Yunfan Lou, Yifan Ye, Yankai Fu, Jun Cen+6

Peking University Beijing, China · Nanjing University Nanjing, China · State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University Beijing, China

Tactile World Model Contact-Rich Dexterous Manipulation Efficient World-Action Model Robotic Manipulation World Models Robot Systems+1

Open paper AskPDF arXiv

World action models inherit the predictive capability of world models, enabling action generation to be guided by anticipated future observations. However, they rely primarily on vision and often fail in contact-rich manipulation, where critical cues arise from physical interaction. In this paper, we propose Dream-Tac, a unified Tactile-World Action Model that jointly models actions, future visual observations, and tactile dynamics. Specifically, Dream-Tac introduces (i) contact-gated visuotactile fusion to selectively integrate tactile signals and (ii) a contact-aware attention bias to better regulate cross-modal interactions during manipulation. To support real-time deployment, we further design a dual-level acceleration strategy, reformulating the contact-aware bias to preserve the fused attention path during training and introducing cache-based diffusion acceleration at inference, achieving up to 2.9

\times

faster training and 1.8

\times

faster inference. Across six contact-rich manipulation tasks, Dream-Tac improves action accuracy by 31.7% on average, demonstrating the effectiveness of unified visuotactile world modeling.Code is available at https://github.com/LYFCLOUDFAN/Dream-Tac.

Abstract

Explore similar work

Dream-Tac: A Unified Tactile World Action Model for Contact-Rich Robot Manipulation

TacForeSight: Force-Guided Tactile World Model for Contact-Rich Manipulation

ReTouch: Empowering Contact-Rich Dexterous Manipulation with Online-Refined Tactile Prediction