{ "schema_version": "paper_public_manifest_v1", "paper_id": "gemini-robotics-2_2026_07", "slug": "gemini-robotics-2", "title": "The overlooked non-consensus is whole-body unified control", "authors": [], "source": { "arxiv_id": "", "pdf_url": "", "project_url": "https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots", "github_url": "", "huggingface_url": "", "original_source": "https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots" }, "site": { "post_url": "/posts/gemini-robotics-2", "canonical_url": "https://haiguangboy.com/posts/gemini-robotics-2", "cover_image": "https://static.haiguangboy.com/papers/gemini-robotics-2/cover.webp" }, "taxonomy": { "domain": "embodied_ai", "track": "vla", "tasks": [ "embodied_ai", "vla", "action_generation", "robotics", "Embodied intelligence", "Robotics", "geminirobotics" ], "related_topics": [ { "paper_id": "latepost_xuhuazhe_202603_2026_03", "title": "latepost_xuhuazhe_202603", "url": "https://haiguangboy.com/posts/latepost_xuhuazhe_202603", "relation": "contrast", "summary": "Architecture: releasing three models at once, layered rather than end-to-end integrated contradicts the 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": "same_track", "summary": "The entire article has almost no verifiable numbers: success rates exist only in charts, not verifiable at the text level validates boundary: no published robot task completion rates/independent contribution shares, currently more like a data expedition than commercial deployment", "strength": "strong" }, { "paper_id": "wx_晚点latepost_20260718_2026_07", "title": "Sudu Technology debuts at WAIC: Reality is the endpoint for robots, not the boundary of training", "url": "https://haiguangboy.com/posts/sudu-third-position", "relation": "same_track", "summary": "The entire article has almost no verifiable numbers: success rates exist only in charts, not verifiable at the text level validates boundary: a single founder interview, with core technology and commercial claims all self-reported; the 20 billion valuation/10,000-card computing power/POC progress have no third-party verification", "strength": "strong" }, { "paper_id": "leapbot_wa_world_anchor_action_models_via_predictive_latent_alignments_2026_07", "title": "Predictive features cannot be directly fed into diffusion models", "url": "https://haiguangboy.com/posts/leapbot-wa", "relation": "same_track", "summary": "The entire article has almost no verifiable numbers: success rates exist only in charts, not verifiable at the text level validates real-machine validation is thin: a single UR5, only pick-and-place of fruits and vegetables, with no success rate numbers or baseline comparisons", "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": "Cross-embodiment costs are publicly priced: a few hours, usually fewer than 200 examples contradicts ★two technical judgments: collection equipment must be 'design for AI' and co-developed with the robot embodiment; all resources focused on the data loop", "strength": "medium" }, { "paper_id": "deepmind_blog_gemini_robotics_er_2_20260731_2026_07", "title": "Gemini Robotics ER 2", "url": "https://haiguangboy.com/posts/deepmind_blog_gemini_robotics_er_2_20260731", "relation": "same_track", "summary": "Access is tiered: only ER 2 is public, VLA and on-device models are limited to early partners validates product surface: ER 2 follows public preview, with capability list including calling Search and Calendar", "strength": "medium" } ] }, "analyst_take": { "type": "author_opinion", "text": "\"Layered or unified\"—this debate may have been asking the wrong question all along. One camp in the industry argues that behavior/action must be a unified model, opposing modular assembly; another splits physical intelligence into upper and lower layers. Google's approach is both: the upper layer is split, with reasoning and execution as two models connected via tool calls; the lower layer is welded shut, with legs and hands not to be separated.\n\nSo the question is not whether to layer, but where to make the cut. Cutting between semantics and physics benefits both sides; cutting within the physical layer—such as assigning walking and manipulation to two separate models—leaves seams exactly where coordination is most needed.\n\nA side note: this judgment did not only emerge after Google's release. A domestic team had already written \"humanoid manipulation cannot be understood as a robotic arm plus a mobile base\" into a complete technical roadmap, more than half a month before this release. I'll break down that paper next.\n\nAs for the cross-embodiment claim, don't jump to conclusions: the official figures of a few hours were only given for dual arms, with no statement on whether whole-body humanoids are the same; other work has also found that raw trajectories struggle to express contact patterns, constraint geometry, and force directions, so transfer at the physical execution layer is far from trivial." }, "ruling": { "importance_score": 3.0, "one_sentence": "The overlooked non-consensus is whole-body unified control" }, "asset_base_url": "https://static.haiguangboy.com/papers/gemini-robotics-2", "assets": [ { "type": "cover_image", "object_key": "papers/gemini-robotics-2/cover.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/gemini-robotics-2/cover.webp", "role": "post_cover", "size_bytes": 86590 }, { "type": "public_brief", "object_key": "papers/gemini-robotics-2/public_brief.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/gemini-robotics-2/public_brief.md", "role": "public_brief", "size_bytes": 4710 }, { "type": "public_manifest", "object_key": "papers/gemini-robotics-2/public_manifest.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/gemini-robotics-2/public_manifest.json", "role": "public_manifest", "size_bytes": 6558 } ], "published_at": "2026-07-31T12:45:53+08:00", "created_at": "2026-07-31T12:45:53+08:00", "updated_at": "2026-09-02T10:55:06+08:00" }