arXiv:2605.27154v1 Announce Type: new 
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.

قدم الباحثون Touch-R1، وهو نموذج جديد للغة متعددة الوسائط يعتمد على التفكير اللمسي، مبني على Qwen2.5-VL-7B، والذي يهدف إلى تعزيز قدرات التفكير اللمسي في الذكاء الاصطناعي. يتم دعم هذا التطور من خلال مجموعة بيانات TouchReason-1M، التي تضم أكثر من مليون زوج لمسي متزامن، وإطار TouchReason-Bench لتقييم الإدراك اللمسي.

Investigadores han presentado Touch-R1, un nuevo modelo de lenguaje multimodal basado en el razonamiento táctil, construido sobre Qwen2.5-VL-7B, que busca mejorar las capacidades de razonamiento táctil en IA. Este desarrollo está respaldado por el conjunto de datos TouchReason-1M, que comprende más de un millón de pares táctiles sincronizados, y el marco TouchReason-Bench para evaluar la percepción táctil.

Des chercheurs ont présenté Touch-R1, un nouveau modèle de langage multimodal basé sur le raisonnement tactile, construit sur Qwen2.5-VL-7B, qui vise à améliorer les capacités de raisonnement tactile en IA. Ce développement est soutenu par le jeu de données TouchReason-1M, comprenant plus d'un million de paires tactiles synchronisées, et le cadre TouchReason-Bench pour évaluer la perception tactile.

Researchers have introduced Touch-R1, a novel tactile reasoning multimodal large language model (MLLM) built on Qwen2.5-VL-7B, which aims to enhance tactile reasoning capabilities in AI. This development is supported by the TouchReason-1M dataset, comprising over 1 million synchronized tactile pairs, and the TouchReason-Bench framework for evaluating tactile perception.

Touch-R1: Reinforcing Touch Reasoning in MLLMs

arXiv:2605.29563v2 Announce Type: replace-cross 
Abstract: Can VLMs predict how each camera move changes the view, and plan many such moves ahead? We call this capability view planning, requiring (1)understanding how a single action transforms the view, and (2)composing many such transformations across multi-turn plans to identify a target view. We probe both abilities in our proposed ViewSuite, a 3D point-cloud environment on real ScanNet scenes. Across 13 frontier VLMs, a critical planning gap emerges: they possess basic view-action knowledge but fail to compose it across multi-turn plans, with the gap widening as viewpoint distance grows. To close this gap, we propose an iterative framework that alternates self-exploration with view graph distillation. The key insight is that all exploration trajectories, regardless of their outcome, collectively form a view graph that compactly captures how viewpoints connect across a scene. Distilling this graph into diverse supervised tasks reshapes the policy distribution and overcomes the sparse rewards that stall pure RL. This improves Qwen2.5-VL-7B from 2.5% to 47.8% on interactive view planning, surpassing GPT-5.4 Pro (18.5%) and Gemini 3.1 Pro (21.4%). Self-exploration emerges as a promising path toward VLMs that can actively reason and plan in 3D space. Code and Data are at https://viewsuite.github.io.

تقدم دراسة حديثة إطارًا لتخطيط المشاهد في نماذج اللغة المرئية (VLM)، مع التركيز على قدرتها على التنبؤ بكيفية تغيير حركات الكاميرا للمشاهد والتخطيط لعدة إجراءات لتحقيق مشهد مستهدف. تبرز الأبحاث فجوة كبيرة في نماذج VLM الحالية، التي يمكنها فهم التحولات الأساسية للمشاهد لكنها تكافح مع الخطط المعقدة متعددة الأدوار، خاصة مع زيادة مسافة نقطة المشاهدة.

Un estudio reciente introduce un marco para la planificación de vistas en Modelos de Lenguaje Visual (VLM), enfatizando su capacidad para predecir cómo los movimientos de cámara alteran las vistas y planificar múltiples acciones para lograr una vista objetivo. La investigación destaca una brecha significativa en los VLM actuales, que pueden entender transformaciones de vista básicas pero luchan con planes complejos de múltiples turnos, especialmente a medida que aumenta la distancia del punto de vista.

Une étude récente présente un cadre pour la planification de vues dans les modèles de langage visuel (VLM), soulignant leur capacité à prédire comment les mouvements de caméra modifient les vues et à planifier plusieurs actions pour atteindre une vue cible. La recherche met en évidence un écart significatif dans les VLM actuels, qui peuvent comprendre des transformations de vue de base mais ont du mal avec des plans multi-tours complexes, en particulier à mesure que les distances de point de vue augmentent.

A recent study introduces a framework for view planning in Vision Language Models (VLMs), emphasizing their ability to predict how camera movements alter views and plan multiple actions to achieve a target view. The research highlights a significant gap in current VLMs, which can understand basic view transformations but struggle with complex multi-turn plans, particularly as viewpoint distances increase.

Touch-R1: Reinforcing Touch Reasoning in MLLMs

Was this article worth reading? Share it

Ready to build your own newsroom?