{ "schema_version": "paper_public_manifest_v1", "paper_id": "agile-wam_2026_09", "slug": "agile-wam", "title": "Agile-WAM：轻量视触觉联合预测", "authors": [], "source": { "arxiv_id": "2609.20761v1", "pdf_url": "https://arxiv.org/pdf/2609.20761v1", "project_url": "", "github_url": "", "huggingface_url": "", "original_source": "https://arxiv.org/pdf/2609.20761v1" }, "site": { "post_url": "/posts/agile-wam", "canonical_url": "https://haiguangboy.com/posts/agile-wam", "cover_image": "https://static.haiguangboy.com/papers/agile-wam/cover.webp" }, "taxonomy": { "domain": "embodied_ai", "track": "world_model", "tasks": [ "embodied_ai", "world_model", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "机器人", "世界模型", "wam", "触觉", "机器人学习" ], "related_topics": [ { "paper_id": "world_action_models_for_robot_learning_and_control_a_survey_2026_09", "title": "WAM综述：预测如何服务控制", "url": "https://haiguangboy.com/posts/wam-survey", "relation": "contrast", "summary": "保留大潜空间但去掉预测损失，收益仍会下降 contradicts 世界预测与动作选择可以分开优化", "strength": "strong" }, { "paper_id": "n_0_vtla_scaling_vision_tactile_language_action_model_with_latent_tactile_tokens_2026_07", "title": "触觉当预测目标而非观测输入", "url": "https://haiguangboy.com/posts/n_0_vtla_scaling_vision_tactile_language_action_model_with_latent_tactile_tokens", "relation": "contrast", "summary": "触觉当预测目标而非观测输入", "strength": "strong" }, { "paper_id": "jepa_wam_learning_vision_language_action_policies_with_joint_embedding_world_modeling_2026_08", "title": "JEPA-WAM：创新方法让世界模型更懂状态变化", "url": "https://haiguangboy.com/posts/jepa-wam", "relation": "same_track", "summary": "保留大潜空间但去掉预测损失，收益仍会下降 validates 五组消融:联合目标 79.2 对未来单独目标 77.3,保空间结构对 iREPA 卷积 74.7", "strength": "strong" }, { "paper_id": "mathcaln_0_foundation_towards_the_age_of_tactile_intelligence_2026_09", "title": "N0-Foundation:开启触觉新时代", "url": "https://haiguangboy.com/posts/mathcaln_0_foundation_towards_the_age_of_tactile_intelligence", "relation": "same_track", "summary": "N0-Foundation:开启触觉新时代", "strength": "strong" }, { "paper_id": "prism_precision_and_contact_rich_real_world_industrial_skill_dataset_with_multim_2026_08", "title": "PRISM:规模能否替代显式建模?", "url": "https://haiguangboy.com/posts/prism_precision_and_contact_rich_real_world_industrial_skill_dataset_with_multim", "relation": "same_track", "summary": "PRISM:规模能否替代显式建模?", "strength": "strong" }, { "paper_id": "smoothrl_online_reinforcement_learning_during_asynchronous_execution_2026_09", "title": "SmoothRL:只对真正执行的动作算账", "url": "https://haiguangboy.com/posts/smoothrl_online_reinforcement_learning_during_asynchronous_execution", "relation": "same_track", "summary": "SmoothRL:只对真正执行的动作算账", "strength": "strong" } ] }, "analyst_take": { "type": "author_opinion", "text": "和T-Rex放在一起看，差别很具体：T-Rex在执行时拆分慢视觉与快触觉修正通路；Agile-WAM在训练时区分视触觉预测时距。两者都重视时间尺度，但“下一步触觉预测”并不等于“高频触觉闭环”。\n工程上值得复核的是：换任务、换采样率后，分开设置预测时距是否仍优于统一时距？这决定了它能否成为可复用的方法，而不只是当前配置下的优势。" }, "ruling": { "importance_score": 3.0, "one_sentence": "Agile-WAM：轻量视触觉联合预测" }, "asset_base_url": "https://static.haiguangboy.com/papers/agile-wam", "assets": [ { "type": "pdf_screenshot", "object_key": "papers/agile-wam/page_01.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/agile-wam/page_01.webp", "role": "paper_first_page", "size_bytes": 205396 }, { "type": "pdf_screenshot", "object_key": "papers/agile-wam/key_figure.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/agile-wam/key_figure.webp", "role": "method_figure", "size_bytes": 262106 }, { "type": "cover_image", "object_key": "papers/agile-wam/cover.webp", "content_type": "image/webp", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/agile-wam/cover.webp", "role": "post_cover", "size_bytes": 69042 }, { "type": "public_brief", "object_key": "papers/agile-wam/public_brief.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/agile-wam/public_brief.md", "role": "public_brief", "size_bytes": 4062 }, { "type": "public_brief", "object_key": "papers/agile-wam/public_brief.en.md", "content_type": "text/markdown; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/agile-wam/public_brief.en.md", "role": "public_brief_en", "size_bytes": 5214 }, { "type": "public_manifest", "object_key": "papers/agile-wam/public_manifest.en.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/agile-wam/public_manifest.en.json", "role": "public_manifest_en", "size_bytes": 6467 }, { "type": "public_manifest", "object_key": "papers/agile-wam/public_manifest.json", "content_type": "application/json; charset=utf-8", "upload_status": "uploaded", "bucket": "paper-assets", "url": "https://static.haiguangboy.com/papers/agile-wam/public_manifest.json", "role": "public_manifest", "size_bytes": 6984 } ], "published_at": "2026-09-21T17:55:00+08:00", "created_at": "2026-09-21T17:55:00+08:00", "updated_at": "2026-09-21T17:55:09+08:00" }