Touch-R1: Reinforcing Touch Reasoning in MLLMs

• URL: http://arxiv.org/abs/2605.27154v1
• Date: Tue, 26 May 2026 15:14:56 GMT
• Title: Touch-R1: Reinforcing Touch Reasoning in MLLMs
• Abstract summary: We introduce TouchReason-1M, a large-scale multimodal dataset, and TouchReason-Bench, a rigorous framework for evaluating tactile perception and conflict resolution.<n>Touch-R1 is trained via a tactile-grounded GRPO objective that combines ordinal-aware accuracy, cross-sensor physical consistency, structured-format control, and an input-side tactile grounding objective.<n>On TouchReason-Bench, Touch-R1-7B outperforms Octopi-13B by 18.4% and GPT-4o by 24.7% on average.
• Score: 7.657861449099524
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: While rule-based reinforcement learning has recently catalyzed explicit reasoning in multimodal models, tactile reasoning remains largely underexplored. Existing tactile-language models primarily rely on supervised or contrastive objectives, which limits their capacity to ground predictions in physical evidence or rectify misleading visual priors. Tactile reasoning introduces two modality-specific challenges: the ordinal nature of physical attributes (e.g., hardness, roughness) and the cross-sensor distribution shifts inherent in optical tactile hardware. In this work, we introduce TouchReason-1M, a large-scale multimodal dataset comprising over 1M synchronized tactile pairs across four distinct sensors, and TouchReason-Bench, a rigorous framework for evaluating tactile perception and visual-tactile conflict resolution. Building upon these, we propose Touch-R1, a tactile reasoning MLLM based on Qwen2.5-VL-7B. Touch-R1 is trained via a tactile-grounded GRPO objective that combines ordinal-aware accuracy, cross-sensor physical consistency, structured-format control, and an input-side tactile grounding objective. Specifically, the tactile-use reward assigns credit only when authentic tactile inputs yield superior correctness relative to counterfactual controls where the tactile stream is removed, shuffled, or noise-masked. On TouchReason-Bench, Touch-R1-7B outperforms Octopi-13B by 18.4\% and GPT-4o by 24.7\% on average. Its structured reasoning traces reveal emergent behaviors of probing, comparison, and revision, demonstrating that R1-style reasoning can be effectively grounded in physical contact.

Related papers

• FELT: Generating Tactile Signals from Vision for Visuo-Tactile Manipulation [19.661583133282477]
  We present Feature-Extracted Latent Tactile (FELT), a learning-based framework that synthesizes per-finger pressure tactile images from RGB observations.<n>FELT uses a large frozen visual encoder and a lightweight query decoder to predict tactile signals in a single feed-forward pass.<n>Experiments on four contact-rich manipulation tasks demonstrate that both generated tactile images and latent tactile features improve policy success.
  arXiv  Detail & Related papers  (2026-07-22T19:32:42Z)
• TacReasoner: A Dynamic Tactile-Language Framework for Interactive Reasoning in Real-World Scenarios [8.86126866988006]
  We propose TacReasoner, a dynamic tactile-language framework for interactive reasoning in real-world scenarios.<n>First, TacReasoner incorporates a Dynamic-aware Tactile to enhance the perception and representation of dynamic tactile signals.<n>We establish DynTac-Bench to systematically evaluate dynamic tactile perception and real-world commonsense reasoning.
  arXiv  Detail & Related papers  (2026-07-06T14:17:44Z)
• Seeing Touch from Motion: A Unified Modality-Aware Visuo-Tactile Policy with Tactile Motion Correlation [74.87716678195099]
  Visuo-Tactile policies leveraging optical tactile sensors have shown great promise in contact-rich manipulation.<n>Despite their advantages, extracting fine-grained contact states necessary for contact-rich manipulation remains an open challenge.<n>We propose a motion-aware tactile representation to facilitate contact-rich manipulation.
  arXiv  Detail & Related papers  (2026-06-29T08:20:18Z)
• Tactile-WAM: Touch-Aware World Action Model with Tactile Asymmetric Attention [53.84395207661823]
  We propose Tactile-WAM, a touch-aware WAM with a Tactile Asymmetric Attention Mechanism (TAAM)<n>On ManiFeel, Tactile-WAM improves the mean success rate by 38.9% overall and by 86% on contact-rich tasks.
  arXiv  Detail & Related papers  (2026-06-25T06:54:56Z)
• FTP-1: A Generalist Foundation Tactile Policy Across Tactile Sensors for Contact-Rich Manipulation [66.22092641373067]
  FTP-1 is the first generalist foundation tactile policy pretrained to acquire transferable tactile manipulation abilities.<n>It supports varied tactile inputs, including image-, array-, and state-based signals.<n>It learns tactile skills that transfer beyond the sensors seen during pretraining.
  arXiv  Detail & Related papers  (2026-06-11T09:30:09Z)
• Blind Dexterous Grasping via Real2Sim2Real Tactile Policy Learning [6.509572742369941]
  We propose a framework for tactile-only blind grasping that is deployable on a physical multi-fingered robotic hand.<n>We introduce a Real2Sim tactile calibration pipeline that constructs a contact-calibrated digital-twin simulator capable of reproducing real tactile signals.<n>Second, we improve the expressiveness of sparse tactile observations using a layout-aware tactile encoder, which incorporates sensor-geometry priors through self-supervised pretraining.<n>Third, to improve generalization to unseen objects, we train object-specific reinforcement-learning experts in the calibrated simulator and aggregate their successful grasp trajectories into a tactile-conditioned Diffusion Policy.
  arXiv  Detail & Related papers  (2026-06-10T07:46:38Z)
• TouchThinker: Scaling Tactile Commonsense Reasoning to the Open World with Large-scale Data and Action-aware Representation [50.608989079323784]
  We propose TouchThinker, a tactile-language framework that scales tactile commonsense reasoning to the open world from both data and representation perspectives.<n>First, we construct TouchThinker-1M, a million-scale, multi-source tactile reasoning dataset covering textbf415 objects, textbf8 scenarios, and textbf7 sensor types.<n>Then, we propose action-aware modeling mechanism to improve tactile representation efficiency and enable efficient reasoning.
  arXiv  Detail & Related papers  (2026-06-10T03:58:32Z)
• FG-CLTP: Fine-Grained Contrastive Language Tactile Pretraining for Robotic Manipulation [10.959531491338552]
  We develop a 3D tactile-language-action (3D-TLA) architecture driven by a flow matching policy to enable reasoning and control.<n>The proposed FG-C model yields a 95.9% classification accuracy and reduces the regression error (MAE) by 52.6%.<n>Building upon this fine-grained representation, we develop a 3D tactile-language-action (3D-TLA) architecture driven by a flow matching policy to enable reasoning and control.
  arXiv  Detail & Related papers  (2026-03-11T15:21:54Z)
• Semantic-Contact Fields for Category-Level Generalizable Tactile Tool Manipulation [82.63833405368159]
  Generalizing tool manipulation requires both semantic planning and precise physical control.<n>We propose Semantic-Contact Fields (SCFields), a unified 3D representation fusing visual semantics with dense contact estimates.<n> Experiments on scraping, crayon drawing, and peeling demonstrate robust category-level generalization.
  arXiv  Detail & Related papers  (2026-02-14T16:05:08Z)
• UniForce: A Unified Latent Force Model for Robot Manipulation with Diverse Tactile Sensors [51.88112610411651]
  We propose a novel unified tactile representation learning framework that learns a shared latent force space across diverse tactile sensors.<n>UniForce reduces cross-sensor domain shift by jointly modeling inverse dynamics (image-to-force) and forward dynamics (force-to-image)<n>To avoid reliance on expensive external force/torque (F/T) sensors, we exploit static equilibrium and collect force-paired data via direct sensor--object--sensor interactions.
  arXiv  Detail & Related papers  (2026-02-01T11:03:01Z)
• TaF-VLA: Tactile-Force Alignment in Vision-Language-Action Models for Force-aware Manipulation [14.094740703476903]
  We introduce TaF-VLA, a framework that grounds high-dimensional tactile observations in physical interaction forces.<n>TaF-VLA policy significantly outperforms state-of-the-art tactile-vision-aligned and vision-only baselines on contact-rich tasks.
  arXiv  Detail & Related papers  (2026-01-28T07:34:41Z)
• TouchGuide: Inference-Time Steering of Visuomotor Policies via Touch Guidance [53.35296919674763]
  TouchGuide is a novel cross-policy visuo-tactile fusion paradigm that fuses modalities within a low-dimensional action space.<n>touchGuide operates in two stages to guide a pre-trained diffusion or flow-matching visuomotor policy.<n>To facilitate TouchGuide training with high-quality and cost-effective data, we introduce TacUMI, a data collection system.
  arXiv  Detail & Related papers  (2026-01-28T04:22:47Z)
• Towards Generalization of Tactile Image Generation: Reference-Free Evaluation in a Leakage-Free Setting [25.355424080824996]
  Tactile sensing is critical for human perception and underpins applications in computer vision, robotics, and multimodal learning.<n>Because tactile data is often scarce and costly to acquire, generating synthetic tactile images provides a scalable solution to augment real-world measurements.<n>We demonstrate that overlapping training and test samples in commonly used datasets inflate performance metrics, obscuring the true generalizability of tactile models.
  arXiv  Detail & Related papers  (2025-03-10T02:37:22Z)
• Elastic Tactile Simulation Towards Tactile-Visual Perception [58.44106915440858]
  We propose Elastic Interaction of Particles (EIP) for tactile simulation. EIP models the tactile sensor as a group of coordinated particles, and the elastic property is applied to regulate the deformation of particles during contact. We further propose a tactile-visual perception network that enables information fusion between tactile data and visual images.
  arXiv  Detail & Related papers  (2021-08-11T03:49:59Z)
• The Feeling of Success: Does Touch Sensing Help Predict Grasp Outcomes? [57.366931129764815]
  We collect more than 9,000 grasping trials using a two-finger gripper equipped with GelSight high-resolution tactile sensors on each finger.<n>Our experimental results indicate that incorporating tactile readings substantially improve grasping performance.
  arXiv  Detail & Related papers  (2017-10-16T05:32:38Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.