{ "schema_version": "paper_public_manifest_v1", "paper_id": "sudu-third-position_2026_07", "slug": "sudu-third-position", "title": "Sudu Technology: The Third Position of a Horizontal Cut", "authors": [], "source": { "arxiv_id": "", "pdf_url": "", "project_url": "https://mp.weixin.qq.com/s/-XEZ4-Q_F2Rev5PnaReA5w", "github_url": "", "huggingface_url": "", "original_source": "https://mp.weixin.qq.com/s/-XEZ4-Q_F2Rev5PnaReA5w" }, "site": { "post_url": "/posts/sudu-third-position", "canonical_url": "https://haiguangboy.com/posts/sudu-third-position", "cover_image": "https://static.haiguangboy.com/papers/sudu-third-position/cover.webp" }, "taxonomy": { "domain": "embodied_ai", "track": "world_model", "tasks": [ "embodied_ai", "world_model", "action_generation", "robotics", "state_prediction", "Embodied Intelligence", "Robotics", "Technical Route", "Sudu Technology" ], "related_topics": [ { "paper_id": "wx_界面新闻_20260605_2026_06", "title": "wx_Interface News_20260605", "url": "https://haiguangboy.com/posts/wx_界面新闻_20260605", "relation": "contrast", "summary": "★Core positioning: Splits general physical intelligence into upper and lower layers; Sudu only handles lower-layer execution and explicitly does not do upper-layer LLM/Agent contradicts route bet: Doing the brain, not the body—betting on the foundational model company identity and rejecting the athletic capability track", "strength": "strong" }, { "paper_id": "wx_中国财富_20260719_2026_07", "title": "Qianxun Intelligence Han Fengtao: Embodied Intelligence Enters the 'Speed Competition' Stage", "url": "https://haiguangboy.com/posts/qianxun-general-brain-data-loop", "relation": "contrast", "summary": "★★Route positioning: Sudu is the third position in the 'horizontal tech stack cut'—neither a general brain nor a vertical industry brain contradicts ★Route bet: Betting on a general brain rather than a vertical industry brain—'In the era of large models, the only opportunity is a general model'", "strength": "strong" }, { "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": "contrast", "summary": "Precise description of the simulation-first training route: Pretraining uses almost no real robot data, but after deployment there is failure retry plus a small amount of real robot RL later contradicts Route bet: Inverted data pyramid—human data as the foundation rather than internet video as the foundation", "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": "★Core positioning: Splits general physical intelligence into upper and lower layers; Sudu only handles lower-layer execution and explicitly does not do upper-layer LLM/Agent validates joint co-training of heterogeneous human-robot data is a structurally suboptimal solution", "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 Tactile Sensing Needs Separate Modeling", "url": "https://haiguangboy.com/posts/t_rex_tactile_reactive_dexterous_manipulation", "relation": "same_track", "summary": "T-Rex: Why Tactile Sensing Needs Separate Modeling", "strength": "strong" } ] }, "ruling": { "importance_score": 3.0, "one_sentence": "Sudu Technology: The Third Position of a Horizontal Cut" }, "asset_base_url": "https://static.haiguangboy.com/papers/sudu-third-position", "assets": [ { "type": "public_brief", "object_key": "papers/sudu-third-position/public_brief.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/sudu-third-position/public_brief.md", "role": "public_brief", "size_bytes": 4565 }, { "type": "public_manifest", "object_key": "papers/sudu-third-position/public_manifest.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/sudu-third-position/public_manifest.json", "role": "public_manifest", "size_bytes": 4966 } ], "published_at": "2026-07-23T09:17:33+08:00", "created_at": "2026-07-23T09:17:33+08:00", "updated_at": "2026-09-02T10:55:06+08:00", "analyst_take": { "type": "author_opinion", "text": "💡Where the real disagreement lies\n\nSudu's simulation route collides hard with several entries in the library: Lingchu's 'human data as the foundation' and Genesis's 'simulation only for evaluation, not training data' are both strong 3/3 clashes.\n\nBut what's more worth noting is its relationship with Xu Huazhe—not opposition, but **limitation**. Xu Huazhe argues that 'the action part must be a unified model, opposing modular splicing.' Sudu doesn't deny this, but locks it into the lower layer: the action layer must be unified internally, but the action layer and the planning layer can—and should—be separated.\n\nThis actually answers a question we've been tracking for a long time: does the value of long-horizon tasks lie in orchestration or in atomic capabilities? Sudu doesn't try to make the scheduling layer smarter—acknowledging that the upper layer can be slow, can call external knowledge, or even have processes manually defined by the factory—and bets everything on the reliability and transferability of atomic skills.\n\nBetting on atomic capabilities. This is a clear stance, and one that can be falsified." } }