🎥 Awesome Touch

A curated, learning-friendly list of tactile sensing research resources for robotic manipulation: papers, datasets, benchmarks, simulation platforms, and frameworks, covering the integration of touch with Vision-Language-Action (VLA) models and World Models for contact-rich manipulation. Time range: 2025.01 – 2026.07.

✅ Inclusion Criteria

• Prioritize high-quality works from top conferences and journals such as ICRA / IROS / CoRL / RSS / NeurIPS / ICML / ICLR / CVPR / ICCV / ECCV
• Cover core directions: Tactile VLA Models, Visuo-Tactile World Models, Tactile World Action Models (WAM), Tactile Imitation Learning, Tactile Simulation & Sim2Real, Tactile Datasets & Benchmarks
• Time range: January 2025 – July 2026

---

Table of Contents

1. Surveys & Overviews
2. Tactile VLA Models
   • 2.1 Unified Vision-Tactile-Language-Action Frameworks
   • 2.2 Tactile-Enhanced VLA with Foundation Models
3. Visuo-Tactile World Models
4. Tactile World Action Models (WAM)
5. Tactile Policy Learning
   • 5.1 Imitation Learning with Tactile Feedback
   • 5.2 Reinforcement Learning with Tactile Sensing
   • 5.3 Diffusion Policies for Tactile Manipulation
6. Tactile Simulation & Sim2Real
7. Tactile Sensors & Hardware
8. Tactile Datasets & Benchmarks
9. Key Research Challenges
10. Contact
11. License

---

📚 Surveys & Overviews

(Sorted by publication date, newest first)

• Sensing the Action: Rethinking Sensor Modalities and Multi-Modal Fusion in Vision–Language–Action Models for Robotic Manipulation

  • Publication: Sensors 2026
  • Highlights: Systematic taxonomy of major sensor modalities including RGB, depth, tactile sensing, force/torque, proprioception and IMU; reinterprets VLA within a sensor–fusion–action pipeline.
  • Paper Link: DOI

• Humanoid Locomotion and Manipulation: Current Progress and Challenges in Control, Planning, and Learning

  • Publication: IEEE/ASME Transactions on Mechatronics 2025
  • Highlights: Comprehensive overview of humanoid locomotion and manipulation; highlights emerging role of tactile sensing, particularly whole-body tactile feedback, as crucial modality for contact-rich interactions.
  • Paper Link: arXiv

• Multimodal Perception-Driven Decision-Making for Human-Robot Interaction: A Survey

  • Publication: Frontiers in Robotics and AI 2025
  • Highlights: Comprehensive review of multimodal perception integrating vision, language, and tactile information for decision-making in robotics.
  • Paper Link: Frontiers

---

🤖 Tactile VLA Models

2.1 Unified Vision-Tactile-Language-Action Frameworks

(Sorted by publication date, newest first)

• TacFiLM: Tactile Modality Fusion for Vision-Language-Action Models

  • Publication: arXiv 2026
  • Highlights: Lightweight modality-fusion approach integrating visual-tactile signals for contact-rich tasks such as insertions, grasping, and in-hand manipulation.
  • Paper Link: arXiv

• AT-VLA: Adaptive Tactile Injection for Enhanced Feedback Reaction in Vision-Language-Action Models

  • Publication: arXiv 2026
  • Highlights: Adaptive tactile injection strategy for enhanced feedback reaction in VLA models; dynamically determines timing and locations for tactile injection.
  • Paper Link: arXiv

• Feeling the Unexpected: ResTacVLA for Contact-Rich Manipulation via Residual Tactile Representation

  • Publication: arXiv 2026 (IROS 2026)
  • Highlights: Addresses modality collapse where high-bandwidth visual features overshadow sparse tactile cues; inspired by Predictive Coding.
  • Paper Link: arXiv

• UniTacVLA: Unified Tactile Understanding and Prediction in Vision Language Action Models

  • Publication: arXiv 2026 (CoRL 2026)
  • Highlights: Unified tactile learning framework modeling tactile signals as dynamic interaction cues for both contact understanding and prediction; introduces tactile-action mixed controller.
  • Paper Link: arXiv

• TaF-VLA: Tactile-Force Alignment in Vision-Language-Action Models for Force-aware Manipulation

  • Publication: arXiv 2026
  • Highlights: Explicitly grounds high-dimensional tactile observations in physical interaction forces.
  • Paper Link: arXiv

• TacMamba: A Tactile History Compression Adapter Bridging Fast Reflexes and Slow VLA Reasoning

  • Publication: arXiv 2026
  • Highlights: Asynchronous architecture: tactile encoder (System 1) performs continuous streaming inference at 100Hz; VLA planner (System 2) queries compressed tactile history on-demand.
  • Paper Link: arXiv

• Modular Sensory Stream for Integrating Physical Feedback in Vision-Language-Action Models (MoSS)

  • Publication: arXiv 2026
  • Highlights: Modular sensory stream for integrating physical feedback in VLA models.
  • Paper Link: arXiv

• MoDE-VLA: Mixture-of-Dexterous-Experts VLA

  • Publication: arXiv 2026
  • Highlights: Seamlessly integrates heterogeneous force and tactile modalities into a pretrained VLA backbone.
  • Paper Link: arXiv

• TacCoRL: Integrating Tactile Feedback into VLA via Simulation

  • Publication: arXiv 2026
  • Highlights: Scalable framework that injects tactile feedback into VLA models via simulation.
  • Paper Link: arXiv

• τ: Learning Touch-Augmented Vision-Language-Action Models from Future Visual Supervision

  • Publication: arXiv 2026
  • Highlights: Learns touch-augmented VLA models from future visual supervision.
  • Paper Link: arXiv

• Tactile-VLA: Unlocking Vision-Language-Action Model's Physical Knowledge for Tactile Generalization

  • Publication: arXiv 2025
  • Highlights: Deeply fuses vision, language, action, and tactile sensing; incorporates hybrid position-force controller and reasoning module for tactile-aware instruction following and zero-shot generalization in contact-rich tasks.
  • Paper Link: arXiv

• OmniVTLA: Vision-Tactile-Language-Action Model with Semantic-Aligned Tactile Sensing

  • Publication: arXiv 2025
  • Highlights: Dual-path tactile encoder with pretrained ViT and semantically-aligned tactile ViT (SA-ViT); introduces ObjTac dataset with 135K tri-modal samples; achieves 96.9% success with grippers and 100% with dexterous hands.
  • Paper Link: arXiv

• VLA-Touch: Enhancing Vision-Language-Action Models with Dual-Level Tactile Feedback

  • Publication: arXiv 2025
  • Highlights: Enhances VLA with tactile sensing without fine-tuning the base VLA; leverages pretrained tactile-language model for semantic feedback and diffusion-based controller for action refinement.
  • Paper Link: arXiv

• VTLA: Vision-Tactile-Language-Action Model with Preference Learning for Insertion Manipulation

  • Publication: arXiv 2025
  • Highlights: Integrates tactile perception into vision-language foundation model for contact-rich manipulation challenges such as peg-in-hole insertion.
  • Paper Link: arXiv

• End-to-End Dexterous Arm-Hand VLA Policies via Shared Autonomy

  • Publication: arXiv 2025
  • Highlights: Shared Autonomy framework where human guides arm pose via VR teleoperation while autonomous DexGrasp-VLA policy handles hand control using real-time tactile and visual feedback.
  • Paper Link: arXiv

• HapticVLA: Contact-Rich Manipulation via Vision-Language-Action Model without Inference-Time Tactile Sensing

  • Publication: arXiv 2025
  • Highlights: Safety-Aware Reward-Weighted Flow Matching (SA-RWFM) and Tactile Distillation (TD) framework incorporating distilled tactile token rather than raw sensor feedback.
  • Paper Link: arXiv

2.2 Tactile-Enhanced VLA with Foundation Models

(Sorted by publication date, newest first)

• DreamTacVLA: Learning to Feel the Future for Contact-Rich Manipulation

  • Publication: arXiv 2025
  • Highlights: Hierarchical perception scheme with macro/local/micro vision; Hierarchical Spatial Alignment (HSA) loss aligns tactile tokens with spatial counterparts; tactile world model predicts future tactile signals; achieves up to 95% success.
  • Paper Link: arXiv

• VLH: Vision-Language-Haptics Foundation Model

  • Publication: arXiv 2025
  • Highlights: Unifies perception, language, and tactile feedback in aerial robotics and VR; processes visual inputs and language instructions via fine-tuned OpenVLA backbone; achieves 56.7% success for target acquisition and 100% accuracy in texture discrimination.
  • Paper Link: arXiv

• Demonstrating the Octopi-1.5 Visual-Tactile-Language Model

  • Publication: arXiv 2025
  • Highlights: Demonstration of Octopi-1.5 with ability to process tactile signals from multiple object parts.
  • Paper Link: arXiv

---

🌍 Visuo-Tactile World Models

(Sorted by publication date, newest first)

• OmniVTA: Visuo-Tactile World Modeling for Contact-Rich Robotic Manipulation

  • Publication: arXiv 2026
  • Highlights: Large-scale OmniViTac dataset with 21,000+ tasks and 86 objects; world-model-based framework with self-supervised tactile encoder, two-stream visuo-tactile world model, contact-aware fusion policy, and 60Hz reflexive controller.
  • Paper Link: arXiv

• ViTacWorld: Scaling Visuo-Tactile World Models for Contact-Rich Robot Manipulation

  • Publication: arXiv 2026
  • Highlights: First framework using world model for robot visuo-tactile-action trajectory generation and policy evaluation; leverages public real tactile datasets and simulation environment for scalable data augmentation.
  • Paper Link: arXiv

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

  • Publication: arXiv 2026
  • Highlights: Lightweight force-conditioned tactile foresight framework; TacForceWM predicts short-horizon tactile latent dynamics from dual-finger tactile observations conditioned on wrist force/torque; enables proactive contact reasoning with real-time inference.
  • Paper Link: arXiv

• TouchWorld: A Predictive and Reactive Tactile Foundation Model for Dexterous Manipulation

  • Publication: arXiv 2026
  • Highlights: Hierarchical policy separating vision-language subtask planning, tactile world-model prediction, visuo-tactile goal-conditioned action generation, and high-frequency tactile residual refinement; achieves 65.0% success across six long-horizon tasks.
  • Paper Link: arXiv

• ContactWorld: What Matters in Vision-Tactile World Models for Contact-Rich Manipulation

  • Publication: arXiv 2026
  • Highlights: Benchmark and systematic empirical study of vision-tactile world models across 12 contact-rich tasks; finds spatially structured and temporally continuous representations achieve strongest planning performance.
  • Paper Link: arXiv

• Visuo-Tactile World Models (VT-WM)

  • Publication: arXiv 2026
  • Highlights: First multi-task visuo-tactile world model capturing physics of contact through touch reasoning; 33% better object permanence, 29% better compliance with laws of motion; zero-shot real-robot experiments achieve up to 35% higher success.
  • Paper Link: arXiv

---

⚡ Tactile World Action Models (WAM)

(Sorted by publication date, newest first)

• 𝒩₀-TWAM: Scaling Tactile-Native World Action Model for Contact-Rich Manipulation

  • Publication: arXiv 2026
  • Highlights: First tactile-native world-action model trained at large scale; predicts both future vision and contact with strong capability on contact-rich tasks.
  • Paper Link: arXiv

• VT-WAM: Visual-Tactile World Action Model for Contact-Rich Manipulation

  • Publication: arXiv 2026
  • Highlights: Jointly learns future visual prediction, tactile deformation prediction, and action prediction within unified flow matching framework; outperforms Fast-WAM by 26.67% and OmniVTLA by 35.84%.
  • Paper Link: arXiv

• Tactile-WAM: Touch-Aware World Action Model with Tactile Asymmetric Attention

  • Publication: arXiv 2026
  • Highlights: Introduces Tactile Asymmetric Attention Mechanism (TAAM) with VideoClean mask to prevent tactile pollution; improves mean success rate by 38.9% overall and 86% on contact-rich tasks.
  • Paper Link: arXiv

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

  • Publication: arXiv 2026
  • Highlights: Unified Tactile-World Action Model jointly modeling actions, future visual observations, and tactile dynamics; contact-gated visuotactile fusion and contact-aware attention bias; improves action accuracy by 31.7%.
  • Paper Link: arXiv

• VTAM: Video-Tactile-Action Models for Complex Physical Interaction Beyond VLAs

  • Publication: arXiv 2026
  • Highlights: Visuo-tactile world action model integrating high-resolution tactile sensing with visual observations within predictive video backbone for robust contact-rich manipulation.
  • Paper Link: arXiv

---

🎯 Tactile Policy Learning

5.1 Imitation Learning with Tactile Feedback

(Sorted by publication date, newest first)

• On the Importance of Tactile Sensing for Imitation Learning: A Case Study on Robotic Match Lighting

  • Publication: IEEE RA-L 2026
  • Highlights: Multimodal visuotactile imitation learning framework with modular transformer architecture and flow-based generative model for fast and dexterous manipulation policies.
  • Paper Link: arXiv

• ViTacFormer: Learning Cross-Modal Representation for Visuo-Tactile Dexterous Manipulation

  • Publication: RSS 2026
  • Highlights: Cross-modal representation learning for visuo-tactile dexterous manipulation.
  • Paper Link: arXiv

• Inference-time Policy Steering via Vision and Touch (ViTaL)

  • Publication: arXiv 2026
  • Highlights: Visuo-tactile inference-time steering framework formulating multimodal guidance as bi-level optimization; improves overall success by 51% over base policy.
  • Paper Link: arXiv

• TouchGuide: Inference-Time Steering of Visuomotor Policies via Touch Guidance

  • Publication: RSS 2026
  • Highlights: Cross-policy visuo-tactile fusion paradigm fusing modalities within low-dimensional action space.
  • Paper Link: arXiv

• Contact-Grounded Policy: Dexterous Visuotactile Policy with Generative Contact Grounding

  • Publication: RSS 2026
  • Highlights: Dexterous visuotactile policy with generative contact grounding.
  • Paper Link: arXiv

• Gentle Object Retraction in Dense Clutter Using Multimodal Force Sensing and Imitation Learning

  • Publication: IEEE RA-L 2026
  • Highlights: Investigates role of contact force sensing for training robots to gently reach into constrained clutter.
  • Paper Link: arXiv

• FBI: Learning Dexterous In-hand Manipulation with Dynamic Visuotactile Shortcut Policy

  • Publication: arXiv 2025
  • Highlights: Flow Before Imitation (FBI) framework dynamically fuses tactile interactions with visual observations through motion dynamics; transformer-based interaction module with one-step diffusion policy for real-time execution.
  • Paper Link: arXiv

• Touch begins where vision ends: Generalizable policies for contact-rich manipulation (ViTaL)

  • Publication: arXiv 2025
  • Highlights: VisuoTactile Local policy learning; removing tactile input reduces success rates by average 40%.
  • Paper Link: arXiv

• Grasp Like Humans: Learning Generalizable Multi-Fingered Grasping from Human Proprioceptive Sensorimotor Integration

  • Publication: IEEE RA-L 2025
  • Highlights: Glove-mediated tactile-kinematic perception-prediction framework for grasp skill transfer from human operation to robotic execution.
  • Paper Link: arXiv

• TACT: Humanoid Whole-body Contact Manipulation through Deep Imitation Learning with Tactile Modality

  • Publication: IEEE RA-L 2025
  • Highlights: Humanoid control system with tactile sensors on upper body for whole-body manipulation via imitation learning from human teleoperation data.
  • Paper Link: arXiv

• Feel the Force: Contact-Driven Learning from Humans (FTF)

  • Publication: arXiv 2025
  • Highlights: Models human tactile-proprioceptive signals during manipulation; trains closed-loop imitation learning policy; achieves 77% success rate across 5 force-sensitive tasks.
  • Paper Link: arXiv

5.2 Reinforcement Learning with Tactile Sensing

(Sorted by publication date, newest first)

• Learning Tactile-Aware Quadrupedal Loco-Manipulation Policies
  • Publication: arXiv 2026
  • Highlights: Hierarchical policy with real-world human demonstrations for tactile-conditioned visuotactile high-level policy; large-scale RL in simulation for tactile-aware whole-body control; zero-shot transfer improves performance by 28.54%.
  • Paper Link: arXiv

5.3 Diffusion Policies for Tactile Manipulation

(Sorted by publication date, newest first)

• Residual Rotation Correction using Tactile Equivariance (EquiTac)

  • Publication: IEEE RA-L 2026
  • Highlights: Exploits SO(2) symmetry of in-hand object rotation for sample-efficient visuotactile policy learning; first tactile learning method to explicitly encode tactile equivariance.
  • Paper Link: arXiv

• PolyTouch: A Robust Multi-Modal Tactile Sensor for Contact-rich Manipulation Using Tactile-Diffusion Policies

  • Publication: ICRA 2025 (Best Paper Nominee)
  • Highlights: Robot finger integrating camera-based tactile sensing, acoustic sensing, and peripheral visual sensing; at least 20-fold increase in lifespan; tactile-diffusion policy significantly outperforms haptic-oblivious policies.
  • Paper Link: arXiv

• VT-Refine: Learning Bimanual Assembly with Visuo-Tactile Feedback via Simulation Fine-Tuning

  • Publication: CoRL 2025
  • Highlights: Combines real-world demonstrations, high-fidelity tactile simulation, and RL for precise bimanual assembly; diffusion policy trained on small demonstrations then refined via large-scale RL in simulated digital twin.
  • Paper Link: arXiv

• Reactive Diffusion Policy: Slow-Fast Visual-Tactile Policy Learning for Contact-Rich Manipulation

  • Publication: RSS 2025
  • Highlights: TactAR teleoperation system with real-time tactile feedback through AR; Reactive Diffusion Policy (RDP) with two-level hierarchy for contact-rich manipulation skills.
  • Paper Link: arXiv

• Visuotactile-Based Learning for Insertion with Compliant Hands

  • Publication: IEEE RA-L 2025
  • Highlights: Simulation-based multimodal policy learning with all-around tactile sensing and depth camera; transformer-based policy via teacher-student distillation; zero-shot transfer.
  • Paper Link: arXiv

---

🧪 Tactile Simulation & Sim2Real

(Sorted by publication date, newest first)

• ETac: A Lightweight and Efficient Tactile Simulation Framework for Learning Dexterous Manipulation

  • Publication: ICRA 2026 (Best Student Paper Award)
  • Highlights: Lightweight and efficient tactile simulation framework for learning dexterous manipulation.
  • Paper Link: arXiv

• Tac2Real: Reliable and GPU Visuotactile Simulation for Online Reinforcement Learning and Zero-Shot Real-World Deployment

  • Publication: arXiv 2026
  • Highlights: GPU visuotactile simulation for online RL and zero-shot real-world deployment.
  • Paper Link: arXiv

• UniVTAC: A Unified Simulation Platform for Visuo-Tactile Manipulation Data Generation, Learning, and Benchmarking

  • Publication: arXiv 2026
  • Highlights: Unified simulation platform with eight representative visuo-tactile manipulation tasks; UniVTAC Encoder improves average success rates by 17.1%.
  • Paper Link: arXiv

• TacSL: A Library for Visuotactile Sensor Simulation and Learning

  • Publication: IEEE TRO 2025
  • Highlights: GPU-based visuotactile sensor simulation over 200× faster than prior SOTA within Isaac Simulator; asymmetric actor-critic distillation (AACD) for sim-to-real transfer.
  • Paper Link: arXiv

• Taccel: Scaling Up Vision-based Tactile Robotics via High-performance GPU Simulation

  • Publication: arXiv 2025
  • Highlights: High-performance simulation platform integrating IPC and ABD to model robots, tactile sensors, and objects; achieves 18-fold acceleration over real-time across thousands of parallel environments.
  • Paper Link: arXiv

• SimTac: A Physics-Based Simulator for Vision-Based Tactile Sensing with Biomorphic Structures

  • Publication: arXiv 2025
  • Highlights: Particle-based deformation modeling, light-field rendering for photorealistic tactile image generation, and neural network for predicting mechanical responses.
  • Paper Link: arXiv

• Tactile MNIST: Benchmarking Active Tactile Perception

  • Publication: arXiv 2025
  • Highlights: Open-source, Gymnasium-compatible benchmark for active tactile perception tasks including localization, classification, and volume.
  • Paper Link: arXiv

• Hydrosoft: Non-Holonomic Hydroelastic Models for Compliant Tactile Manipulation

  • Publication: arXiv 2025
  • Highlights: Computationally efficient non-holonomic hydroelastic model for path-dependent contact force distributions and dynamic surface area variations.
  • Paper Link: arXiv

• Closing the Reality Gap: Zero-Shot Sim-to-Real Deployment for Dexterous Force-Based Grasping and Manipulation

  • Publication: arXiv 2026
  • Highlights: Holistic sim-to-real recipe combining scalable tactile simulation, current-to-torque calibration, and actuator modelling for dexterous manipulation.
  • Paper Link: arXiv

---

🔧 Tactile Sensors & Hardware

(Sorted by publication date, newest first)

• ArrayTac: A Closed-loop Piezoelectric Tactile Platform for Continuously Tunable Rendering of Shape, Stiffness, and Friction

  • Publication: arXiv 2026
  • Highlights: Piezoelectric-driven tactile display capable of simultaneously rendering shape, stiffness, and friction to reproduce realistic haptic signals.
  • Paper Link: arXiv

• T-Rex: Tactile-Reactive Dexterous Manipulation

  • Publication: arXiv 2026
  • Highlights: Variable-rate Mixture-of-Transformers (MoT) architecture with novel temporal tactile VQ-VAE encoder for high-frequency touch signals.
  • Paper Link: arXiv

• PolyTouch: A Robust Multi-Modal Tactile Sensor

  • Publication: ICRA 2025
  • Highlights: Novel robot finger integrating camera-based tactile sensing, acoustic sensing, and peripheral visual sensing; compact, durable, easy-to-manufacture; at least 20× lifespan increase.
  • Paper Link: arXiv

• OSMO: Open-Source Tactile Glove for Human-to-Robot Skill Transfer

  • Publication: arXiv 2025
  • Highlights: 12 three-axis tactile sensors across fingertips and palm; compatible with state-of-the-art hand-tracking methods for in-the-wild data collection; achieves 72% success rate on wiping task.
  • Paper Link: arXiv

• TwinTac: A Wide-Range, Highly Sensitive Tactile Sensor with Real-to-Sim Digital Twin Sensor Model

  • Publication: arXiv 2025
  • Highlights: Hardware sensor designed for high sensitivity and wide measurement range; digital twin model for simulation.
  • Paper Link: arXiv

• FreeTacMan: Robot-free Visuo-Tactile Data Collection System for Contact-rich Manipulation

  • Publication: arXiv 2025
  • Highlights: Human-centric, robot-free data collection system leveraging dexterity and force feedback of human motion; comprises 10k demonstration trajectories across 50 tasks.
  • Paper Link: arXiv

---

📊 Tactile Datasets & Benchmarks

(Sorted by publication date, newest first)

Dataset/Benchmark Name	Year	Scale	Core Tasks	Link
OmniViTac	2026	21,000+ tasks, 86 objects	Visuo-tactile-action dataset with six physics-grounded interaction patterns	arXiv
ContactWorld	2026	12 contact-rich tasks	Vision-tactile world model benchmark: insertion, disassembly, screwing, exploratory interaction	arXiv
ManiFeel	2026	Contact-rich tasks	Benchmark for tactile world action models	arXiv
WorldArena 2.0	2026	Extended benchmark	Extends benchmarking from visual to visuotactile world models	arXiv
UniVTAC Benchmark	2026	8 visuo-tactile tasks	Evaluating tactile-driven policies	arXiv
ObjTac	2025	135K tri-modal samples, 56 objects, 10 categories	Force-based tactile dataset with textual, visual, and tactile information	arXiv
Tactile MNIST	2025	Active tactile perception	Localization, classification, and volume tasks	arXiv
ManiSkill-ViTac 2025 Challenge	2025	Unified platform	Vision and tactile fusion in robotic manipulation	Website

---

🔑 Key Research Challenges

Based on synthesis of 2025–2026 literature, key research challenges in tactile robotic manipulation include:

• Tactile Pollution — Unconstrained tactile-token injection can degrade video and action prediction by forcing visual dynamics models to absorb sparse, local, event-driven contact signals.

• Modality Collapse — High-bandwidth visual features can overshadow sparse tactile cues in VLA models, requiring careful architectural design.

• Tactile Data Scarcity — Real tactile interaction data are expensive to collect, hardware-dependent, and limited in task and scene diversity.

• Sim-to-Real Gap for Tactile — Tactile signals are directly grounded in physical contact and can exhibit a smaller simulation-to-real gap than visual observations, but simulation remains challenging.

• Cross-Modal Representation Compatibility — Effectiveness of tactile sensing depends critically on cross-modal representation compatibility rather than modality scaling alone.

• Spatial-Temporal Structure — Representations that are both spatially structured and temporally continuous consistently achieve the strongest planning performance in contact-rich settings.

• Contact-Centric Fidelity — Contact-centric fidelity (event sequence and local dynamics) is a necessary condition for task success in contact-rich manipulation.

• Real-Time Inference — Tactile feedback loops require high-frequency processing (e.g., 60Hz) for closed-loop control, demanding efficient world model inference.

---

❤️ Contact

If you have suggestions for new papers or datasets, spot any inaccuracies, or find this curated resource list helpful, feel free to reach out to Yuanliang Sun via email at sun254667307@gmail.com. For inquiries about research internship opportunities in egocentric vision, you are welcome to contact me at any time.

📄 License

CC0 1.0 Universal. Full terms available in the LICENSE file.