{ "schema_version": "paper_public_manifest_v1", "paper_id": "t_rex_tactile_reactive_dexterous_manipulation_2026_07", "slug": "t_rex_tactile_reactive_dexterous_manipulation", "title": "T-Rex: Why Tactile Sensing Needs Its Own Model", "authors": [], "source": { "arxiv_id": "2606.17055v1", "pdf_url": "https://arxiv.org/pdf/2606.17055v1", "project_url": "", "github_url": "", "huggingface_url": "", "original_source": "https://arxiv.org/abs/2606.17055v1" }, "site": { "post_url": "/posts/t_rex_tactile_reactive_dexterous_manipulation", "canonical_url": "https://haiguangboy.com/posts/t_rex_tactile_reactive_dexterous_manipulation", "cover_image": "https://static.haiguangboy.com/papers/t_rex_tactile_reactive_dexterous_manipulation/cover.webp" }, "taxonomy": { "domain": "embodied_ai", "track": "vla", "tasks": [ "embodied_ai", "vla", "action_generation", "robotics", "latent_state", "state_prediction", "Embodied Intelligence", "Tactile Sensing", "VLA Architecture", "Dexterous Manipulation", "Robot Learning" ], "related_topics": [ { "paper_id": "an_open_foundation_model_towards_2026_07", "title": "An_Open_Foundation_Model_Towards", "url": "https://haiguangboy.com/posts/an_open_foundation_model_towards", "relation": "same_track", "summary": "Three-stage training recipe: large-scale human egocentric pretraining → tactile grounding training → skill post-training validates that joint co-training on heterogeneous human-robot data is structurally suboptimal", "strength": "strong" }, { "paper_id": "wx_界面新闻_20260605_2026_06", "title": "wx_界面新闻_20260605", "url": "https://haiguangboy.com/posts/wx_界面新闻_20260605", "relation": "contrast", "summary": "Route assessment: tactile signals differ fundamentally from vision and require dedicated asynchronous architectures; naive fusion is ineffective contradicts architectural bet: unified networks outperform modular pipelines—WUM opposes VLA's layer-by-layer semantic transfer", "strength": "medium" }, { "paper_id": "latepost_xuhuazhe_202603_2026_03", "title": "latepost_xuhuazhe_202603", "url": "https://haiguangboy.com/posts/latepost_xuhuazhe_202603", "relation": "same_track", "summary": "Boundary: single robot platform + lab-collected data + self-reported evaluation, including one proactively disclosed failure case validates boundary: single media interview self-report, no third-party verification, founder admits path uncertainty", "strength": "medium" }, { "paper_id": "sunday_blog_20260717_2026_07", "title": "sunday_blog_20260717", "url": "https://haiguangboy.com/posts/sunday_blog_20260717", "relation": "same_track", "summary": "Boundary: single robot platform + lab-collected data + self-reported evaluation, including one proactively disclosed failure case validates source boundary: self-reported, self-built evaluation, no third-party, single task family", "strength": "medium" }, { "paper_id": "orca_2026_07", "title": "π0.5 Shivers in Place After Failing to Grab a Spoon, While Orca Goes Further with Physical Intuition Learned from Watching Videos", "url": "https://haiguangboy.com/posts/orca", "relation": "same_track", "summary": "The key to a world model is a readable state", "strength": "medium" }, { "paper_id": "wx_星河频率_20260718_2026_07", "title": "Robots Begin to 'Stand in the Light': Lingchu Intelligence Enters Optical Module Production Lines", "url": "https://haiguangboy.com/posts/lingchu-optical", "relation": "same_track", "summary": "Route alignment: the three-stage recipe embodies the philosophy of 'massive non-action data pretraining + small action data alignment' validates the 'native human data' pyramid claim: pretraining relies primarily on human data, with real-robot data used only for post-training adaptation", "strength": "medium" } ] }, "ruling": { "importance_score": 3.0, "one_sentence": "T-Rex: Why Tactile Sensing Needs Its Own Model" }, "asset_base_url": "https://static.haiguangboy.com/papers/t_rex_tactile_reactive_dexterous_manipulation", "assets": [ { "type": "cover_image", "object_key": "papers/t_rex_tactile_reactive_dexterous_manipulation/cover.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/t_rex_tactile_reactive_dexterous_manipulation/cover.webp", "role": "post_cover", "size_bytes": 87876 }, { "type": "public_brief", "object_key": "papers/t_rex_tactile_reactive_dexterous_manipulation/public_brief.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/t_rex_tactile_reactive_dexterous_manipulation/public_brief.md", "role": "public_brief", "size_bytes": 4495 }, { "type": "public_manifest", "object_key": "papers/t_rex_tactile_reactive_dexterous_manipulation/public_manifest.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/t_rex_tactile_reactive_dexterous_manipulation/public_manifest.json", "role": "public_manifest", "size_bytes": 5548 } ], "published_at": "2026-07-21T17:41:47+08:00", "created_at": "2026-07-21T17:41:47+08:00", "updated_at": "2026-09-02T10:55:06+08:00", "analyst_take": { "type": "author_opinion", "text": "💡 The Real Architectural Bet\n\nA unified robot system does not mean every modality must be squeezed into the same encoder, running at the same frequency with the same representation.\n\nA more likely architecture: each physical signal first forms a specialized neural pathway according to its own nature, then coordinates at the action-decision level or in a shared latent space. Robots can have a unified brain, but that does not mean they can only have one kind of sensory pathway.\n\nGoing deeper, T-Rex's three-stage training echoes another recurring route in the knowledge base: massive human video first builds physical and motor priors, while a small amount of robot data handles final alignment. What is truly scarce may not just be action labels, but how to convert larger-scale 'non-action data' into executable capability." } }