{ "schema_version": "paper_public_manifest_v1", "paper_id": "leapbot-wa_2026_07", "slug": "leapbot-wa", "title": "Predictive features cannot be directly fed into diffusion models", "authors": [], "source": { "arxiv_id": "2607.23969", "pdf_url": "https://arxiv.org/pdf/2607.23969", "project_url": "", "github_url": "", "huggingface_url": "", "original_source": "https://arxiv.org/abs/2607.23969" }, "site": { "post_url": "/posts/leapbot-wa", "canonical_url": "https://haiguangboy.com/posts/leapbot-wa", "cover_image": "https://static.haiguangboy.com/papers/leapbot-wa/cover.webp" }, "taxonomy": { "domain": "embodied_ai", "track": "world_model", "tasks": [ "embodied_ai", "world_model", "action_generation", "robotics", "latent_state", "state_prediction", "Embodied intelligence", "World models", "jepa" ], "related_topics": [ { "paper_id": "robointer15_a_holistic_intermediate_representation_suite_for_embodied_world_mode_2026_07", "title": "It's not \"whether to have a world model,\" it's \"whether what it outputs has structure\"", "url": "https://haiguangboy.com/posts/robointer15_a_holistic_intermediate_representation_suite_for_embodied_world_mode", "relation": "contrast", "summary": "It's not \"whether to have a world model,\" it's \"whether what it outputs has structure\"", "strength": "strong" }, { "paper_id": "fast-wam_2026_07", "title": "No need to imagine the future at inference time—robots still hit 91.8%! Fast-WAM debunks WAM's core assumption", "url": "https://haiguangboy.com/posts/fast-wam", "relation": "contrast", "summary": "Training on video objectives matters more than imagining the future at test time", "strength": "strong" }, { "paper_id": "wx_top华人科创社_20260721_2026_07", "title": "Xu Huazhe welcomes first commercial partner, former Meituan drone executive joins", "url": "https://haiguangboy.com/posts/wx_top华人科创社_20260721", "relation": "contrast", "summary": "Key design: world model only looks at intent, not actions—deliberately strips low-level motor commands to prevent collapsing into a control shortcut contradicts world model architecture: video generation model + prediction node dual model, with action injection/bootstrapped/persistent memory mechanisms", "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" }, { "paper_id": "t_rex_tactile_reactive_dexterous_manipulation_2026_07", "title": "T-Rex: Why touch needs separate modeling", "url": "https://haiguangboy.com/posts/t_rex_tactile_reactive_dexterous_manipulation", "relation": "same_track", "summary": "T-Rex: Why touch needs separate modeling", "strength": "strong" }, { "paper_id": "orca_2026_07", "title": "π0.5 shudders in place when failing to grab a spoon, but Orca goes further with physics intuition learned from watching videos", "url": "https://haiguangboy.com/posts/orca", "relation": "same_track", "summary": "The key to a world model is readable state", "strength": "strong" } ] }, "ruling": { "importance_score": 3.0, "one_sentence": "Predictive features cannot be directly fed into diffusion models" }, "asset_base_url": "https://static.haiguangboy.com/papers/leapbot-wa", "assets": [ { "type": "cover_image", "object_key": "papers/leapbot-wa/cover.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/leapbot-wa/cover.webp", "role": "post_cover", "size_bytes": 82054 }, { "type": "public_brief", "object_key": "papers/leapbot-wa/public_brief.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/leapbot-wa/public_brief.md", "role": "public_brief", "size_bytes": 5081 }, { "type": "public_manifest", "object_key": "papers/leapbot-wa/public_manifest.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/leapbot-wa/public_manifest.json", "role": "public_manifest", "size_bytes": 4892 } ], "published_at": "2026-07-31T06:32:33+08:00", "created_at": "2026-07-31T06:32:33+08:00", "updated_at": "2026-09-02T10:55:06+08:00", "analyst_take": { "type": "author_opinion", "text": "\"Auxiliary signals cannot be plugged into the backbone without constraints—they need a structurally restricted position\"—this idea has been recurring recently in very different directions: some make it a residual term, some a prediction target, some add sparse routing to a dedicated pathway, and this paper uses attention masks to control who sees whom. Touch, sensor calibration, world models—the problem domains differ widely, yet the solutions converge to the same shape.\n\nWhat remains undecided are two other points. First, whether to go through pixels: another study found that \"rendering values as images is more effective than injecting coordinates with an MLP,\" citing the ability to leverage pretrained visual priors; this paper argues for leaving pixels entirely. Both sides have data, but in opposite directions.\n\nSecond, when imagination should occur. This paper prunes the world model branch at deployment, while navigation-oriented work follows a \"think before acting\" approach. However, that line of work itself admits the gains mainly appear in scenarios with high occlusion and dynamic density. So pruning for manipulation is reasonable, keeping it for navigation is reasonable—the divergence may not be about architectural correctness, but about whether task uncertainty is high enough." } }