{ "schema_version": "paper_public_manifest_v1", "paper_id": "towards-agentic-robots_2026_10", "slug": "towards-agentic-robots", "title": "机器人Agent综述：四大方向一篇梳理", "authors": [], "source": { "arxiv_id": "2610.20442", "pdf_url": "https://arxiv.org/pdf/2610.20442", "project_url": "", "github_url": "", "huggingface_url": "", "original_source": "https://arxiv.org/pdf/2610.20442" }, "site": { "post_url": "/posts/towards-agentic-robots", "canonical_url": "https://haiguangboy.com/posts/towards-agentic-robots", "cover_image": "https://static.haiguangboy.com/papers/towards-agentic-robots/cover.webp" }, "taxonomy": { "domain": "embodied_ai", "track": "world_model", "tasks": [ "embodied_ai", "world_model", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "机器人", "agent", "世界模型" ], "related_topics": [ { "paper_id": "learning_a_thousand_tasks_in_a_day_2026_08", "title": "1 天 1000 任务，靠的是归纳偏置", "url": "https://haiguangboy.com/posts/mt3-thousand-tasks", "relation": "same_track", "summary": "修复方式应对应失败来源 validates 失败归因:检索 22.3%、位姿估计 23.9%、分割 19.5%、被抓物体 20.2%", "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": "Agentic不自动意味着更高自治等级 validates ASIMOV-Agentic 的定位:只测高层语义编排,明确不测功能安全架构", "strength": "strong" }, { "paper_id": "google_blog_gemini_robotics_er_2_20260730_2026_07", "title": "Introducing Gemini Robotics ER 2", "url": "https://haiguangboy.com/posts/google_blog_gemini_robotics_er_2_20260730", "relation": "same_track", "summary": "Agentic不自动意味着更高自治等级 validates 架构表述:VLA 模型和导航 API 被并列声明为 agent 的 tool", "strength": "strong" }, { "paper_id": "pave_predictive_alignment_and_value_guided_evolution_for_world_action_policies_2026_09", "title": "PAVE:失败数据也能喂饱策略", "url": "https://haiguangboy.com/posts/pave_predictive_alignment_and_value_guided_evolution_for_world_action_policies", "relation": "same_track", "summary": "PAVE:失败数据也能喂饱策略", "strength": "strong" }, { "paper_id": "pointworld_scaling_3d_world_models_for_in_the_wild_robotic_manipulation_2026_08", "title": "PointWorld:点流统一状态动作", "url": "https://haiguangboy.com/posts/pointworld_scaling_3d_world_models_for_in_the_wild_robotic_manipulation", "relation": "same_track", "summary": "PointWorld:点流统一状态动作", "strength": "strong" }, { "paper_id": "rl_100_performant_robotic_manipulation_with_real_world_reinforcement_learning_2026_08", "title": "RL 不该从零学，该做后训练", "url": "https://haiguangboy.com/posts/rl-100", "relation": "same_track", "summary": "技能演化包含晋升、验证和弃用机制 validates 离线阶段用近似模型 Q 函数当门控,只有预测确有改进才推进行为策略,否则拒绝该次更新", "strength": "strong" } ] }, "analyst_take": { "type": "author_opinion", "text": "💡 串起热点，比孤立追论文更有价值\nChunkTrust关注何时重新获取反馈；RoboHarn-Evo关注如何核验结果、修订经验。放回这张地图，就能看到它们分别在补执行调度和经验复用的环节。\n\n对工程师而言，这篇的收益是：先定位瓶颈属于哪一层，再沿对应方向找方法；评估组合方案时，把成功率、时间、计算和重试预算一起比较。" }, "ruling": { "importance_score": 3.0, "one_sentence": "机器人Agent综述：四大方向一篇梳理" }, "asset_base_url": "https://static.haiguangboy.com/papers/towards-agentic-robots", "assets": [ { "type": "pdf_screenshot", "object_key": "papers/towards-agentic-robots/page_01.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/towards-agentic-robots/page_01.webp", "role": "paper_first_page", "size_bytes": 178322 }, { "type": "pdf_screenshot", "object_key": "papers/towards-agentic-robots/key_figure.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/towards-agentic-robots/key_figure.webp", "role": "method_figure", "size_bytes": 219218 }, { "type": "cover_image", "object_key": "papers/towards-agentic-robots/cover.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/towards-agentic-robots/cover.webp", "role": "post_cover", "size_bytes": 69540 }, { "type": "public_brief", "object_key": "papers/towards-agentic-robots/public_brief.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/towards-agentic-robots/public_brief.md", "role": "public_brief", "size_bytes": 3440 }, { "type": "public_brief", "object_key": "papers/towards-agentic-robots/public_brief.en.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/towards-agentic-robots/public_brief.en.md", "role": "public_brief_en", "size_bytes": 4345 }, { "type": "public_manifest", "object_key": "papers/towards-agentic-robots/public_manifest.en.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/towards-agentic-robots/public_manifest.en.json", "role": "public_manifest_en", "size_bytes": 6583 }, { "type": "public_manifest", "object_key": "papers/towards-agentic-robots/public_manifest.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/towards-agentic-robots/public_manifest.json", "role": "public_manifest", "size_bytes": 7160 } ], "published_at": "2026-10-16T06:47:11+08:00", "created_at": "2026-10-16T06:47:11+08:00", "updated_at": "2026-10-16T06:47:23+08:00" }