{ "schema_version": "paper_public_manifest_v1", "paper_id": "wam-survey_2026_09", "slug": "wam-survey", "title": "WAM综述：预测如何服务控制", "authors": [], "source": { "arxiv_id": "2609.16074", "pdf_url": "https://arxiv.org/pdf/2609.16074", "project_url": "", "github_url": "", "huggingface_url": "", "original_source": "https://arxiv.org/pdf/2609.16074" }, "site": { "post_url": "/posts/wam-survey", "canonical_url": "https://haiguangboy.com/posts/wam-survey", "cover_image": "https://static.haiguangboy.com/papers/wam-survey/cover.webp" }, "taxonomy": { "domain": "embodied_ai", "track": "world_model", "tasks": [ "embodied_ai", "world_model", "action_generation", "robotics", "state_prediction", "具身智能", "机器人", "世界模型", "wam", "机器人学习", "表征学习" ], "related_topics": [ { "paper_id": "openwam_an_open_modular_exploration_towards_systematic_world_action_model_pretra_2026_09", "title": "OpenWAM:正确答案会随规模变", "url": "https://haiguangboy.com/posts/openwam_an_open_modular_exploration_towards_systematic_world_action_model_pretra", "relation": "contrast", "summary": "OpenWAM:正确答案会随规模变", "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": "contrast", "summary": "PAVE:失败数据也能喂饱策略", "strength": "strong" }, { "paper_id": "latepost_xuhuazhe_202603_2026_03", "title": "latepost_xuhuazhe_202603", "url": "https://haiguangboy.com/posts/latepost_xuhuazhe_202603", "relation": "contrast", "summary": "世界预测与动作选择可以分开优化 contradicts 路线赌注：行为/动作部分必须是统一模型，反对模块化拼接", "strength": "strong" }, { "paper_id": "omega_eva_envision_verify_and_act_with_latent_interactive_world_models_2026_08", "title": "世界模型：推理时到底要不要想象", "url": "https://haiguangboy.com/posts/omega-eva", "relation": "contrast", "summary": "世界预测与动作选择可以分开优化 contradicts 核心主张:世界模型该做的不是训练时打辅助,也不是当外部模拟器,而是在动作生成这一步内部当验证器", "strength": "strong" }, { "paper_id": "slim_05b_learning_action_grounded_predictive_latents_for_robot_manipulation_2026_08", "title": "预测误差更低，表征却已经塌了", "url": "https://haiguangboy.com/posts/slim-05b", "relation": "contrast", "summary": "神经模拟器可支持闭环策略改进 contradicts 核心主张:未来预测应当是塑造表征的训练信号,不是模型的输出", "strength": "strong" }, { "paper_id": "d4rt_efficiently_reconstructing_dynamic_scenes_one_2026_06", "title": "D4RT_Efficiently_Reconstructing_Dynamic_Scenes_One", "url": "https://haiguangboy.com/posts/d4rt_efficiently_reconstructing_dynamic_scenes_one", "relation": "same_track", "summary": "二维视频预测不自动保证三维与多视角一致 validates D4RT输出是相对/up-to-scale,不是metric——这是当具身3D数据源的命门", "strength": "strong" } ] }, "analyst_take": { "type": "author_opinion", "text": "💡 和OpenWAM放在一起看\nOpenWAM在其评测中发现：没有具身预训练时，单向信息流略占优；加入预训练后，双向互通的优势才显现。这给综述的分类补上一个条件：架构优劣会随预训练条件改变，不能只凭“统一”或“解耦”选边。\n\n半年后回看，值得检验的是：固定数据和算力预算，分开优化世界与策略，能否减少模型偏差引起的真实执行退化。这个结果，才有助于判断分离优化的价值。" }, "ruling": { "importance_score": 3.0, "one_sentence": "WAM综述：预测如何服务控制" }, "asset_base_url": "https://static.haiguangboy.com/papers/wam-survey", "assets": [ { "type": "pdf_screenshot", "object_key": "papers/wam-survey/page_01.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/wam-survey/page_01.webp", "role": "paper_first_page", "size_bytes": 260458 }, { "type": "pdf_screenshot", "object_key": "papers/wam-survey/key_figure.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/wam-survey/key_figure.webp", "role": "method_figure", "size_bytes": 221648 }, { "type": "cover_image", "object_key": "papers/wam-survey/cover.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/wam-survey/cover.webp", "role": "post_cover", "size_bytes": 63078 }, { "type": "public_brief", "object_key": "papers/wam-survey/public_brief.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/wam-survey/public_brief.md", "role": "public_brief", "size_bytes": 4245 }, { "type": "public_brief", "object_key": "papers/wam-survey/public_brief.en.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/wam-survey/public_brief.en.md", "role": "public_brief_en", "size_bytes": 5216 }, { "type": "public_manifest", "object_key": "papers/wam-survey/public_manifest.en.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/wam-survey/public_manifest.en.json", "role": "public_manifest_en", "size_bytes": 6545 }, { "type": "public_manifest", "object_key": "papers/wam-survey/public_manifest.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/wam-survey/public_manifest.json", "role": "public_manifest", "size_bytes": 7158 } ], "published_at": "2026-09-18T08:26:27+08:00", "created_at": "2026-09-18T08:26:27+08:00", "updated_at": "2026-09-18T08:26:37+08:00" }