Visible Touch: Rendering Contact for Visuomotor Policies

Metin Alp Dogan; Edward Sun; Feng Xu; Daniel Wu; Allen Peng; Dennis Hong; Yuchen Cui

cs.ROSep 12, 2026

Visible Touch: Rendering Contact for Visuomotor Policies

Authors: Metin Alp Dogan, Edward Sun, Feng Xu, Daniel Wu, Allen Peng, Dennis Hong, Yuchen Cui

Organizations: University of California, Los Angeles

Abstract

Integrating contact information into visuomotor policies remains an open problem. Touch is essential to robust manipulation, yet most modern policies, including pretrained vision-language-action (VLA) models, operate from vision and proprioception alone. Existing approaches to closing this gap require specialized tactile hardware, add separate tactile encoders, or commit to non-image policy backbones, all incompatible with the modern paradigm of image-conditioned policies built on pretrained 2D visual representations. Our key insight is that the bottleneck is not the contact information itself, but how it is delivered: when contact signals are exposed in the same spatial frame as the scene the policy already attends to, they become directly usable by any image-conditioned policy without architectural changes. We operationalize this insight in Visible Touch, paired with a custom low-cost magnetic contact sensor that is open-sourced and fabricated from off-the-shelf parts via a parametric CAD-to-mold pipeline. Across the LIBERO benchmark, Visible Touch improves BC-Transformer success by 15.7 percentage points on average in the 2-view setting, with similar gains in the 1-view setting; controlled comparisons show that the contact-integration strategy strongly affects how effectively tactile information is used. The pattern holds when fine-tuning pretrained VLAs: miniVLA on LIBERO gains 25 percentage points on average, and $π_{0.5}$ on four real-world contact-rich tasks gains 30 percentage points with our custom sensor.

Figures & tables

Figure 1: Overview of Visible Touch . A low-cost magnetic contact sensor captures per-taxel contact, which we render as an image-space overlay on the policy’s RGB observations. The augmented images are drop-in inputs for fine-tuning pretrained VLAs, requiring no architectural changes. Contact overlays improve average task success by $+25.3pp$ on LIBERO fine-tuning suites and $+30.3pp$ on real-world contact-rich manipulation tasks.

Figure 2: Hardware system. (a) Visuo-tactile setup: UFactory xArm7 with two RealSense cameras and two tactile sensors mounted on the gripper fingers. (b) Custom tactile sensor with anti-slip skin. (c) Internal structure of the tactile sensor. (d) Our parametric CAD-to-mold pipeline: CAD-generated alternative casting molds (top) and the 3D-printed molds with the fabricated sensor pads (bottom). (e) Physical characteristics of the sensor.

Figure 3: BC-Transformer on LIBERO across camera configurations ( $n{=}3$ seeds, error bars are std). Orange bars show Visible Touch (ours); gray bars show plain baselines; blue bars show the Proprio baseline that receives an aggregated 12-D contact wrench through the proprioceptive state. Visible Touch consistently improves over plain images in both 1-view and 2-view, with the largest gain on LIBERO_LONG ( $+33.7$ pp 2-view). See details in Appendix C .

	Pretraining	Fine-tuning
	LIBERO_90	OBJECT	GOAL	SPATIAL	LONG	Avg.
Baseline	48.3%	51.0%	48.5%	71.5%	32.5%	50.9%
Visible Touch	63.2%	$\mathbf{70.6\pm 4.6}$ %	$\mathbf{80.5\pm 4.3}$ %	$\mathbf{87.0\pm 1.8}$ %	$\mathbf{66.6\pm 2.2}$ %	76.2%

Table 1: miniVLA success rates (%) on LIBERO across pretraining and fine-tuning. LIBERO_90 pretraining results are from single runs. For fine-tuning, baseline entries are the strongest single-run baselines available per suite, while Visible Touch results are mean

\pm

s.e. over 5 training seeds using the suite-specific headline recipe.

Visible Touch variant	OBJECT	GOAL	SPATIAL	LONG	Avg.
Multi-arrows	$94.3\pm 5.6$	$91.8\pm 0.5$	$90.2\pm 1.0$	$76.2\pm 2.7$	$88.1$
Avg. Arrow	$\mathbf{98.2\pm 1.2}$	$\mathbf{92.8\pm 2.5}$	$92.3\pm 2.7$	$\mathbf{86.5\pm 2.3}$	$\mathbf{92.5}$
Binbars	$98.0\pm 0.9$	$92.7\pm 1.7$	$\mathbf{94.5\pm 1.5}$	$78.8\pm 2.6$	$91.0$

Table 2: Success rates (%) for overlay variants on LIBERO suites (BC-Transformer, 2-view,

n{=}3

seeds).

Panel of Figure 4 — Figure 4: Success rates of Visible Touch (Avg. Arrow overlay) over plain baselines, bucketed by gripper-close events. Tasks with no grasps show no benefit; single-grasp tasks show modest gains; multi-grasp tasks show the largest gains. The pattern holds across both BC-Transformer ( 4(a) ) and a pretrained miniVLA on LIBERO_90 ( 4(b) ).

Panel of Figure 5 — Figure 5: Real-world experimental setup. (a) Contact-rich manipulation tasks: Lift, Transfer Tube, Put Mug in Dishwasher , and Plug Charger . (b) Example contact overlays rendered onto external and wrist camera views.

Figure 6: Tac-View : contact arrows as a separate image.

Figure 7: Real-world evaluation of

\pi_{0.5}

fine-tuning across four contact-rich manipulation tasks (successes out of 30 trials per condition, sub-task scored). Final-stage success corresponds to complete end-to-end task success. Hatched bars denote the binary ablations of the matching solid variant: Binary Contact (1x) and (9x) keep only per-finger and per-taxel contact occurrence, respectively. Exact counts are listed in Table 15 ( Section F.3 ).

Appendix figures & tables19 assets

Supplementary material from the paper’s appendix.

Appendix

Sensor pad (consumable)
Component	Spec	Qty	Unit price	Subtotal
XP-565 silicone elastomer	10:1 base/activator	10 g	$0.05/g	$0.50
Neodymium magnets	$\varnothing$ 2 mm $\times$ 1 mm	9	$0.07/ea	$0.63
PLA filament (molds + spacer)	—	20 g	$0.013/g	$0.26
Loctite SF 770	—	1 ml	$0.50/ml	$0.50
Loctite 406	—	0.5 ml	$1.65/ml	$0.83

Appendix

Table 3: Bill of materials for one tactile sensor (

3{\times}3

configuration). Sensor pad components are consumable and can be replaced without replacing the PCB. PCB costs are rough estimates and vary significantly with order quantity.

Figure 8: Fabrication steps of our custom tactile sensor.

Figure 9: Sensor characterization experiments. (a) Experiment setup: UFactory xArm7 equipped with BOTA Force-Torque sensor and our custom tactile sensor, used to collect ground-truth contact normal force data for characterizing the tactile sensor’s response. (b) Standard deviation of the 4 independently fabricated sensor readings under different contact normal forces.

Panel of Figure 10 — Figure 10: Directional shear response and hysteresis. Left: in-plane flux change of the center taxel during $\pm 1$ mm tangential sweeps along X and Y at $0.6$ mm indentation (3 repetitions): orthogonal, monotonic, sign-symmetric response trajectories. Right: aggregate response $r$ versus normal force over six consecutive load–unload cycles at $0.5$ mm/s; maximum loading–unloading separation is $\leq$ 14% of full scale and identical across cycles.

Figure 11: Baseline stability over a 30-minute stationary hold. Top: aggregate response over the full session; orange lines mark the reference presses bracketing the hold. Bottom: signed per-channel baseline deviation (60 s means) of all 27 channels during the hold; 26 of 27 channels stay within $\pm 20\,\mu$ T, and the largest excursion ( $-55\,\mu$ T, center-taxel normal axis) is a settling transient of $0.7\%$ of a typical press signal.

Architecture
Component / Hyperparameter	Value
Visual backbone	ResNet-18 (random init, remove_layer_num=4 , no stride change)
Language fusion	FiLM conditioning on visual features
Language encoder	1-layer MLP (input $768$ , hidden $128$ , output $128$ )
Transformer layers	$4$
Hidden / embed dimension	$64$ (token embed); MLP hidden $256$

Appendix

Table 4: BC-Transformer training configuration for LIBERO multitask experiments.

	1-view (agentview)		2-view (agentview + wrist)
Suite	Plain	Visible Touch	Plain	Proprio	Visible Touch
OBJECT	$90.7\pm 4.4$	$\mathbf{95.0\pm 0.9}$	$82.8\pm 7.0$	$80.0\pm 2.0$	$\mathbf{98.2\pm 1.2}$
GOAL	$79.3\pm 3.2$	$\mathbf{91.7\pm 0.3}$	$86.8\pm 4.1$	$81.5\pm 4.3$	$\mathbf{92.8\pm 2.5}$
SPATIAL	$79.0\pm 2.0$	$\mathbf{90.5\pm 0.4}$	$84.8\pm 4.0$	$80.0\pm 1.4$	$\mathbf{92.3\pm 2.7}$
LONG	$41.8\pm 6.1$	$\mathbf{63.8\pm 1.2}$	$52.8\pm 2.1$	$42.8\pm 5.1$	$\mathbf{86.5\pm 2.3}$
Average	$72.7$	$\mathbf{85.2}$	$76.8$	$71.1$	$\mathbf{92.5}$

Appendix

Table 5: BC-Transformer success rates (%) on LIBERO suites across camera configurations (

n{=}3

seeds).

Suite	Task	Plain 2v	Visible Touch 2v	$\Delta(pp)$
OBJECT	butter $\to$ basket	$48\%$	$98\%$	$+50$
GOAL	wine $\to$ top of cabinet	$83\%$	$100\%$	$+17$
SPATIAL	bowl on ramekin $\to$ plate	$62\%$	$93\%$	$+32$
LONG	cheese + butter $\to$ basket	$8\%$	$95\%$	$+87$

Appendix

Table 6: Rescue outliers: tasks where Avg. Arrow produces the largest gain over plain 2-view within each suite. All four outliers involve grasp configurations that are visually ambiguous (slippery, small, occluded, or perched objects), where the contact arrow provides information not available from pixels alone.

	Raw Pearson		Partial (controlling for headroom)
Feature	$r$	$p$	$r$	$p$
Headroom	$+0.924$	$<10^{-3}$	—	—
#grasps	$+0.633$	$<10^{-3}$	$-0.094$	$0.562$
Trajectory length	$+0.458$	$0.003$	$-0.368$	$0.019$
%contact	$+0.360$	$0.022$	$+0.214$	$0.186$
#contact onsets	$+0.329$	$0.038$	$-0.571$	$<10^{-3}$

Appendix

Table 7: Per-task correlations between

\Delta=

Visible Touch 2v

-

plain 2v and task properties, pooled across all 40 LIBERO tasks. Raw Pearson correlations are dominated by headroom (overlays help most where vision fails). After controlling for headroom via partial correlation, only #contact onsets and trajectory length retain independent predictive signal.

Suite	1-view Plain	2-view Plain	$\Delta$ (2v $-$ 1v)
OBJECT	$90.7$	$82.8$	$\mathbf{-7.9}$
GOAL	$79.3$	$86.8$	$+7.5$
SPATIAL	$79.0$	$84.8$	$+5.8$
LONG	$41.8$	$52.8$	$+11.0$
Average	$72.7$	$76.8$	$+4.1$

Appendix

Table 8: Effect of camera configuration on the plain (no-overlay) BC-Transformer baseline. Mean success rate (%) over 3 seeds per cell, 10 tasks per suite. Three of four suites benefit from the wrist camera; OBJECT regresses.

Suite	Seed 12345	Seed 23456	Seed 34567	Mean	SD
OBJECT	$84.5$	$73.5$	$90.5$	$82.8$	$7.0$
GOAL	$90.0$	$89.5$	$81.0$	$86.8$	$4.1$
SPATIAL	$90.5$	$82.0$	$82.0$	$84.8$	$4.0$
LONG	$50.5$	$52.5$	$55.5$	$52.8$	$2.1$

Appendix

Table 9: Per-seed success rates (%) for the Plain 2-view BC-Transformer baseline.

Suite	Seed 12345	Seed 23456	Seed 34567	Mean	SD
OBJECT	$82.5$	$77.5$	$80.0$	$80.0$	$2.0$
GOAL	$84.0$	$75.5$	$85.0$	$81.5$	$4.3$
SPATIAL	$82.0$	$79.0$	$79.0$	$80.0$	$1.4$
LONG	$50.0$	$38.5$	$40.0$	$42.8$	$5.1$

Appendix

Table 10: Per-seed success rates (%) for the Proprio 2-view baseline (Plain RGB + 12-D contact wrench concatenated to the proprioceptive state).

Suite	Seed 12345	Seed 23456	Seed 34567	Mean	SD
OBJECT	$99.0$	$99.0$	$96.5$	$98.2$	$1.2$
GOAL	$96.0$	$92.5$	$90.0$	$92.8$	$2.5$
SPATIAL	$96.0$	$91.5$	$89.5$	$92.3$	$2.7$
LONG	$84.0$	$86.0$	$89.5$	$86.5$	$2.3$

Appendix

Table 11: Per-seed success rates (%) for Visible Touch 2-view (default Avg. Arrow overlay).

Figure 12: Per-task success rates on LIBERO_90: Visible Touch -miniVLA (contact-overlay pretraining) vs. the re-evaluated Stanford miniVLA [ 1 ] baseline.

Suite	Recipe	Baseline	Visible Touch	$\Delta$
OBJECT	b8 1-epoch	$47.0\%$	$57.5\%$	$+10.5$ pp
OBJECT	DDP-4 b16 cosine	$51.0\%$	$70.6\pm 4.6\%$	$+19.6$ pp
GOAL	b8 1-epoch	$48.5\%$	$74.0\%$	$+25.5$ pp
GOAL	DDP-4 b16 cosine	$37.4\%$	$80.5\pm 4.3\%$	$+43.1$ pp
SPATIAL	2-phase	$71.5\%$	$87.0\pm 1.8\%$	$+15.5$ pp
SPATIAL	DDP-4 b16 cosine	$55.0\%$	$87.5\%$	$+32.5$ pp

Appendix

Table 12: Per-recipe comparison for History-2 miniVLA on LIBERO suites. Baseline : clean non-contact fine-tuning with clean evaluation. Visible Touch : contact-overlay fine-tuning with overlay-augmented evaluation. For the suite-specific recipes used in the main results, Visible Touch performance is reported as mean

\pm

standard error over 5 training seeds; all other entries are single runs. Baseline results are single runs.

Setting	Value
Base checkpoint	pi05_droid
Fine-tuning method	LoRA (backbone $r{=}16$ , action expert $r{=}32$ )
LoRA targets	attention and FFN projections
Image resolution	$224\times 224\times 3$ , 3 cameras
Action horizon	10 steps (1.0 s at 10 Hz)
Batch size	8

Appendix

Table 13:

\pi_{0.5}

LoRA fine-tuning hyperparameters. All 32 runs share these settings; only the input variant differs.

Task	$n$	Length (steps)	Duration (s)	Min/Max (steps)
cube	100	$131\pm 86$	$13.1\pm 8.6$	90 / 972
tube	97	$190\pm 14$	$19.0\pm 1.4$	166 / 238
charger	100	$238\pm 28$	$23.8\pm 2.7$	182 / 329
dishwasher	100	$281\pm 32$	$28.1\pm 3.2$	236 / 554
All	397	$\mathbf{210\pm 74}$	$\mathbf{21.0\pm 7.4}$	90 / 972

Appendix

Table 14: Per-task demonstration statistics (post-acceptance, post-idle-trim).

n

is the number of episodes used in training; length is reported as mean

\pm

standard deviation in control steps and seconds. Min/max bracket the full distribution.

		Lift	Transfer Tube		Put Mug in Dishwasher			Plug Charger
Method		Pick	Pick	Insert	Pull	Pick	Place	Pick	Insert
Baseline		20/30	5/30	0/30	21/30	11/30	11/30	9/30	0/30
Tac-View		16/30	9/30	0/30	22/30	18/30	16/30	17/30	0/30
Position-Only		22/30	21/30	7/30	17/30	14/30	11/30	9/30	0/30
Visible Touch	Binbars	23/30	10/30	1/30	24/30	15/30	15/30	9/30	0/30
	Binary Contact (1x)	19/30	17/30	2/30	13/30	5/30	4/30	16/30	0/30

Appendix

Table 15: Real-world evaluation of

\pi_{0.5}

fine-tuning across four contact-rich manipulation tasks (30 trials per condition, sub-task scored). Final-stage success corresponds to complete end-to-end task success. Binary Contact (1x) and (9x) denote aggregate per-finger and per-taxel binary contact, respectively.

Task	Step budget	Action scale
Lift	150	0.8
Transfer Tube	400	0.4
Put Mug in Dishwasher	600	0.8
Plug Charger	600	0.6

Appendix

Table 16: Rollout budget and action scale per real-world task.

Explore similar work

Aug 2, 2026cs.RO

CAAT: Contact-Aware Attention Scaling and Tactile Masking for Data-Efficient Contact-Rich Manipulation

In contact-rich manipulation, visual observations primarily guide motion in free space, whereas tactile observations become particularly informative during contact. However, standard Transformer-based visuo-tactile policies typically rely on either token concatenation or learnable gating. These approaches lack explicit contact-aware priors, making it difficult to efficiently learn effective cross-modal representations from demonstrations. To address this limitation, we propose CAAT, a lightweight contact-aware framework that explicitly incorporates contact priors through attention scaling and dynamic tactile masking. Specifically, CAAT emphasizes visual information before contact and tactile information during contact. It also suppresses static background tokens by comparing the current tactile observation with a non-contact reference. CAAT can be integrated into commonly used Transformer-based policies without modifying their action decoders. In simulation, integrating CAAT with ACT improves the average success rate by 18.0 percentage points over direct visuo-tactile fusion and by 10.0 percentage points over gated fusion. In real-world experiments using a visuo-tactile UMI platform, CAAT achieves an average success rate of 60.0% across ACT, Diffusion Policy, and

π_0

, outperforming the strongest baseline by an average of 21.1 percentage points. These results demonstrate that explicit contact priors and dynamic tactile masking are effective in improving visuo-tactile policy learning and task performance of diverse policy architectures. https://mrjiangjm.github.io/caat/

Jiaming Jiang, Yuzhe Huang, Hao Liang +7

ShanghaiTech University · Beijing Institute for General Artificial Intelligence · Beihang University +2

Tactile Contact-Rich Manipulation Contact-Rich Dexterous Manipulation Prior Knowledge Integration

Ask PDF

Sep 14, 2026cs.RO

PredTac: Learning Contact-Rich Manipulation with Predicted Touch

Contact-rich manipulation benefits from tactile feedback, yet physical tactile sensors introduce hardware, calibration, synchronization, and maintenance costs that complicate policy learning and deployment. We formulate predicted touch as an alternative to measured tactile input and present PredTac, a framework that learns to infer tactile states from causal visual observations and robot states and uses the predicted touch as an explicit interface for policy learning and execution. A tactile predictor is first trained with tactile supervision and then used to provide contact information without requiring measured tactile input during downstream policy training or execution. We evaluate PredTac across three contact-rich manipulation tasks in simulation and on a real robot, and further examine how policy performance depends on the predicted contact content. In simulation goal-offset evaluations, predicted-touch policies achieve 27.0%, 52.0%, and 44.7% success on USB, Barbed-spike, and Valve, respectively, improving over the visual baseline by 8.0-13.7 percentage points. On the real robot, predicted-touch ACT achieves 70.0%, 50.0%, and 90.0% success on USB insertion, Barbed extraction, and Valve rotation, respectively, with a three-task mean of 70.0%, approaching measured-touch ACT at 72.2% and substantially outperforming visual ACT at 21.1%. Fixed-policy interventions further show that performance is sensitive to the spatial structure of predicted contact, with spatial rearrangement at fixed value distributions reducing Valve success by 10.7 percentage points. These results demonstrate that predicted touch can provide useful contact information for contact-rich manipulation without requiring tactile sensing as a policy input.

Weijia Fan, Daqiang Guo

The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China.

Tactile Contact-Rich Manipulation Robotic Systems

Ask PDF

Jul 26, 2026cs.RO

$N_0$ -VTLA: Scaling Vision-Tactile-Language-Action Model with Latent Tactile Tokens

We present

N_0

-VTLA, a vision-tactile-language-action (VTLA) foundation model capable of (1) fine-grained contact-rich manipulation with tactile perception and tactile-feedback control, and (2) offline policy improvement from stored deployment data. Building on current vision-based backbones, we propose a training recipe for tactile integration consisting of visuo-tactile pre-training, staged tactile-pathway integration, and advantage-conditioned offline policy improvement. During pre-training, the policy learns broad contact priors from NeoData, our large-scale visuo-tactile robot dataset; to our knowledge,

N_0

-VTLA is the first VTLA model pretrained on tactile data at scale. During post-training, we augment the policy with a predictive tactile pathway that distills the contact patterns learned at scale into the fine motion adjustments required by downstream tactile-centric manipulation. For offline policy improvement, we introduce ALTER, an advantage-conditioned offline reinforcement learning method that converts relative progress and trajectory-event comparisons into binary advantage labels for policy training on a fixed deployment corpus, further improving task-specific learning on contact-rich skills such as deformable object manipulation. Across contact-rich benchmarks,

N_0

-VTLA outperforms strong baselines by wide margins: it wins all nine real-robot NeoReal tasks and reaches 63.8% mean success on a twenty-task simulation suite, against 44.0% for the strongest baseline.

N_0

-VTLA policies trained with ALTER reach 75-95% success on three long-horizon real-robot tasks. These results lay a foundation for versatile tactile-driven manipulation policies.

NeoteAI Team, Fudan TEAI Team

NeoteAI Team · Fudan TEAI Team

Tactile

Ask PDF

Abstract

Explore similar work

CAAT: Contact-Aware Attention Scaling and Tactile Masking for Data-Efficient Contact-Rich Manipulation

PredTac: Learning Contact-Rich Manipulation with Predicted Touch

N0N_0N0​-VTLA: Scaling Vision-Tactile-Language-Action Model with Latent Tactile Tokens

$N_0$ -VTLA: Scaling Vision-Tactile-Language-Action Model with Latent Tactile Tokens