{ "schema_version": "paper_public_manifest_v1", "paper_id": "dr-lfd_2026_08", "slug": "dr-lfd", "title": "50 demos achieve 100%, others with 1000 only get 45%—the difference is it removes 'when to switch tactics' from the strategy", "authors": [], "source": { "arxiv_id": "2607.25397", "pdf_url": "https://arxiv.org/pdf/2607.25397", "project_url": "", "github_url": "", "huggingface_url": "", "original_source": "https://arxiv.org/pdf/2607.25397" }, "site": { "post_url": "/posts/dr-lfd", "canonical_url": "https://haiguangboy.com/posts/dr-lfd", "cover_image": "https://static.haiguangboy.com/papers/dr-lfd/cover.webp" }, "taxonomy": { "domain": "embodied_ai", "track": "action_generation", "tasks": [ "embodied_ai", "action_generation", "robotics", "state_prediction", "Embodied intelligence", "Robot learning", "tamp" ], "related_topics": [ { "paper_id": "latepost_xuhuazhe_202603_2026_03", "title": "latepost_xuhuazhe_202603", "url": "https://haiguangboy.com/posts/latepost_xuhuazhe_202603", "relation": "contrast", "summary": "Problem diagnosis: End-to-end policies are forced to learn both 'how to do' and 'when to switch', data requirements explode combinatorially contradicts route bet: behavior/action parts must be a unified model, opposing modular assembly", "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": "Core claim: Switching logic should be left to the planner rather than learned by the policy—this is a structural rebuttal to 'end-to-end scaling' contradicts architecture bet: unified networks outperform modular assembly—WUM opposes VLA's semantic layer-by-layer transmission", "strength": "strong" }, { "paper_id": "n_0_vtla_scaling_vision_tactile_language_action_model_with_latent_tactile_tokens_2026_07", "title": "Tactile sensing as prediction target rather than observation input", "url": "https://haiguangboy.com/posts/n_0_vtla_scaling_vision_tactile_language_action_model_with_latent_tactile_tokens", "relation": "contrast", "summary": "Tactile sensing as prediction target rather than observation input", "strength": "strong" }, { "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": "Problem diagnosis: End-to-end policies are forced to learn both 'how to do' and 'when to switch', data requirements explode combinatorially validates joint co-training of heterogeneous human-robot data is a structurally suboptimal approach", "strength": "strong" }, { "paper_id": "gemini_robotics_2_safety_2026_07", "title": "Gemini_Robotics_2_Safety", "url": "https://haiguangboy.com/posts/gemini_robotics_2_safety", "relation": "same_track", "summary": "Premise: Symbolic planning and visual-motor policies are complementary parallel lines validates empirical finding: there is a clear gap between a model 'understanding constraints' and 'acting according to constraints'", "strength": "strong" }, { "paper_id": "latent_action_pretraining_through_world_modeling_2026_07", "title": "LAWM: Why action labels become a burden", "url": "https://haiguangboy.com/posts/latent_action_pretraining_through_world_modeling", "relation": "same_track", "summary": "LAWM: Why action labels become a burden", "strength": "strong" } ] }, "analyst_take": { "type": "author_opinion", "text": "The most transferable part of this paper is not the architecture but the mechanism: rather than making the policy generalize out-of-distribution, first push the world back into the policy's distribution. Primitives move objects to in-distribution contact states, key poses bring the arm to starting postures the policy can accept. What changes is not the model's generalization ability but the state it is invoked in.\n\nLooking further, recent works give different answers to the same problem. Some advocate unified models against modular assembly; some oppose horizontal cuts like 'navigation module plus grasping module' and favor vertical cuts by abstraction level; this one says cutting is fine, but someone must be responsible for pushing the state back into distribution.\n\nThere is also a direct opposition worth noting: one tactile work's deployment contract is that once an action chunk is issued, it must be executed fully without truncation or re-planning; this paper is the opposite—re-perceive and locally re-solve before every action. One bets predictions are accurate enough so don't interrupt; the other bets the world will drift so calibrate every step. This divergence should show results within six months." }, "ruling": { "importance_score": 3.0, "one_sentence": "Switching logic should not be learned by the policy" }, "asset_base_url": "https://static.haiguangboy.com/papers/dr-lfd", "assets": [ { "type": "pdf_screenshot", "object_key": "papers/dr-lfd/page_01.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/dr-lfd/page_01.webp", "role": "paper_first_page", "size_bytes": 240128 }, { "type": "pdf_screenshot", "object_key": "papers/dr-lfd/key_figure.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/dr-lfd/key_figure.webp", "role": "method_figure", "size_bytes": 276146 }, { "type": "cover_image", "object_key": "papers/dr-lfd/cover.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/dr-lfd/cover.webp", "role": "post_cover", "size_bytes": 81308 }, { "type": "public_brief", "object_key": "papers/dr-lfd/public_brief.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/dr-lfd/public_brief.md", "role": "public_brief", "size_bytes": 4559 }, { "type": "public_manifest", "object_key": "papers/dr-lfd/public_manifest.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/dr-lfd/public_manifest.json", "role": "public_manifest", "size_bytes": 6757 } ], "published_at": "2026-08-01T23:08:28+08:00", "created_at": "2026-08-01T23:08:28+08:00", "updated_at": "2026-09-02T10:55:06+08:00" }