fix(robotdaily): 修复 fallback 模式下的标签和概括生成

arxiv-digest/scripts/enrich_papers.py

@@ -13,7 +13,7 @@ from utils import log, normalize_space, ollama_generate_json, read_json, truncat
 FALLBACK_TAGS = {
     "embodied": ["具身智能", "机器人", "真实部署", "操控", "导航"],
     "representation": ["表征学习", "潜在空间", "世界模型", "预训练", "对象中心"],
-    "reinforcement": ["强化学习", "策略优化", "奖励设计", "离线RL", "模仿学习"],
+    "reinforcement": ["强化学习", "策略优化", "奖励设计", "离线 RL", "模仿学习"],
 }
 
 
@@ -26,7 +26,7 @@ def build_prompt(paper: Dict[str, Any]) -> str:
 {{
   "translated_abstract_zh": "...",
   "brief_explanation_zh": "...",
-  "tags": ["标签1", "标签2", "标签3", "标签4", "标签5"]
+  "tags": ["标签 1", "标签 2", "标签 3", "标签 4", "标签 5"]
 }}
 
 要求：
@@ -43,14 +43,61 @@ def build_prompt(paper: Dict[str, Any]) -> str:
 
 
 def fallback_enrichment(paper: Dict[str, Any]) -> Dict[str, Any]:
-    tags = FALLBACK_TAGS.get(paper["domain"], ["AI论文", "机器学习", "应用研究"])
-    matched = paper.get("matched_applied_terms", [])[:2] + paper.get("matched_innovation_terms", [])[:2]
-    reason = paper.get("selection_reason", "偏应用且具创新性")
-    if matched:
-        reason = f"关键词命中 {', '.join(matched)}，{reason}"
+    tags = FALLBACK_TAGS.get(paper["domain"], ["AI 论文", "机器学习", "应用研究"])
+    title = paper.get("title", "")
+    summary = paper.get("summary", "")
+    
+    # 从标题提取核心方法
+    brief = title.split(':')[0].strip()
+    if '.' in brief:
+        brief = brief.split('.')[0].strip()
+    
+    # 判断方法类型
+    summary_lower = summary.lower()
+    if "diffusion" in summary_lower:
+        method = "扩散模型"
+    elif "reinforcement learning" in summary_lower:
+        method = "强化学习"
+    elif "imitation learning" in summary_lower:
+        method = "模仿学习"
+    elif "contrastive" in summary_lower:
+        method = "对比学习"
+    elif "transformer" in summary_lower:
+        method = "Transformer"
+    elif "self-supervised" in summary_lower:
+        method = "自监督学习"
+    elif "representation" in summary_lower:
+        method = "表征学习"
+    else:
+        method = "多种技术"
+    
+    # 判断应用领域
+    if "robot" in summary_lower or "manipulation" in summary_lower:
+        field = "机器人操作"
+    elif "navigation" in summary_lower or "driving" in summary_lower:
+        field = "导航控制"
+    elif "translation" in summary_lower or "generation" in summary_lower:
+        field = "生成任务"
+    else:
+        field = "相关任务"
+    
+    # 判断结果
+    if "real-world" in summary_lower or "deployment" in summary_lower:
+        result = "真实部署"
+    elif "zero-shot" in summary_lower:
+        result = "零样本泛化"
+    elif "first" in paper.get("matched_innovation_terms", []) or "novel" in paper.get("matched_innovation_terms", []):
+        result = "首次提出"
+    elif "improve" in summary_lower or "better" in summary_lower:
+        result = "性能提升"
+    else:
+        result = "性能优化"
+    
+    brief = f"{brief}，采用{method}解决{field}，实现{result}"
+    
     return {
-        "translated_abstract_zh": f"【LLM 暂不可用，先保留英文摘要要点】{truncate(paper.get('summary', ''), 220)}",
-        "brief_explanation_zh": truncate(reason, 86),
+        "translated_abstract_zh": f"【LLM 暂不可用，先保留英文摘要要点】{truncate(summary, 220)}",
+        "brief_explanation_zh": truncate(brief, 86),
         "tags": tags[:5],
     }
 

arxiv-digest/scripts/run_daily.py

@@ -78,11 +78,16 @@ def main() -> None:
 
     models = parse_models(args.models or env.get("INSIGHT_MODELS", "qwen3.5:cloud,glm-4.7:cloud"))
     if args.skip_enrich:
+        # 使用 fallback 模式：应用预设标签，保留英文摘要要点
+        from enrich_papers import fallback_enrichment
+
         enriched = {k: v for k, v in selection.items() if k != "candidates"}
         for paper in enriched.get("papers", []):
-            paper.setdefault("translated_abstract_zh", paper.get("summary", ""))
-            paper.setdefault("brief_explanation_zh", paper.get("selection_reason", ""))
-            paper.setdefault("tags", [])
+            fallback = fallback_enrichment(paper)
+            paper["translated_abstract_zh"] = fallback["translated_abstract_zh"]
+            paper["brief_explanation_zh"] = fallback["brief_explanation_zh"]
+            paper["tags"] = fallback["tags"]
+            paper["enrichment_model"] = "fallback"
     else:
         enriched = enrich_selection({k: v for k, v in selection.items() if k != "candidates"}, model_names=models)
     write_json(paths["enriched_json"], enriched)

site/content/ai-daily/2026-03-16.md

@@ -1,6 +1,6 @@
 ---
 title: "2026-03-16 · AI 每日简报"
-date: 2026-03-16T16:15:32.720890+08:00
+date: 2026-03-16T16:31:30.888993+08:00
 draft: false
 summary: "RobotDaily 2026-03-16：共 7 篇，含 具身智能 2 篇，表征学习 3 篇，强化学习 2 篇。"
 tags: ["robotdaily", "ai-daily", "具身智能", "表征学习", "强化学习"]
@@ -15,54 +15,54 @@ tags: ["robotdaily", "ai-daily", "具身智能", "表征学习", "强化学习"]
 ## 具身智能（2 篇）
 
 ### 1. PhysMoDPO: Physically-Plausible Humanoid Motion with Preference Optimization
-> 应用信号: real-world, deployment, robot；创新信号: diffusion, zero-shot；领域匹配: robot, humanoid, physical
+> PhysMoDPO，采用扩散模型解决机器人操作，实现真实部署
 - 作者：Yangsong Zhang, Anujith Muraleedharan, Rikhat Akizhanov, Abdul Ahad Butt 等另外4人
-- 标签：—
-- 中文摘要：Recent progress in text-conditioned human motion generation has been largely driven by diffusion models trained on large-scale human motion data. Building on this progress, recent methods attempt to transfer such models for character animation and real robot control by applying a Whole-Body Controller (WBC) that converts diffusion-generated motions into executable trajectories. While WBC trajectories become compliant with physics, they may expos…
+- 标签：`具身智能` `机器人` `真实部署` `操控`
+- 中文摘要：【LLM 暂不可用，先保留英文摘要要点】Recent progress in text-conditioned human motion generation has been largely driven by diffusion models trained on large-scale human motion data. Building on this progress, recent methods attempt to transfer such models…
 - 链接：[DOI](https://doi.org/10.48550/arXiv.2603.13228) | [arXiv](https://arxiv.org/abs/2603.13228v1) | [PDF](https://arxiv.org/pdf/2603.13228v1)
 
 ### 2. HandelBot: Real-World Piano Playing via Fast Adaptation of Dexterous Robot Policies
-> 应用信号: real-world, deployment, robot；创新信号: first；领域匹配: robot, robotics, manipulation
+> HandelBot，采用强化学习解决机器人操作，实现真实部署
 - 作者：Amber Xie, Haozhi Qi, Dorsa Sadigh
-- 标签：—
-- 中文摘要：Mastering dexterous manipulation with multi-fingered hands has been a grand challenge in robotics for decades. Despite its potential, the difficulty of collecting high-quality data remains a primary bottleneck for high-precision tasks. While reinforcement learning and simulation-to-real-world transfer offer a promising alternative, the transferred policies often fail for tasks demanding millimeter-scale precision, such as bimanual piano playing.…
+- 标签：`具身智能` `机器人` `真实部署` `操控`
+- 中文摘要：【LLM 暂不可用，先保留英文摘要要点】Mastering dexterous manipulation with multi-fingered hands has been a grand challenge in robotics for decades. Despite its potential, the difficulty of collecting high-quality data remains a primary bottleneck for high-…
 - 链接：[DOI](https://doi.org/10.48550/arXiv.2603.12243) | [arXiv](https://arxiv.org/abs/2603.12243v1) | [PDF](https://arxiv.org/pdf/2603.12243v1)
 
 ## 表征学习（3 篇）
 
 ### 1. Representation Learning for Spatiotemporal Physical Systems
-> 应用信号: system；创新信号: self-supervised；领域匹配: representation, representations, latent
+> Representation Learning for Spatiotemporal Physical Systems，采用自监督学习解决相关任务，实现性能优化
 - 作者：Helen Qu, Rudy Morel, Michael McCabe, Alberto Bietti 等另外3人
-- 标签：—
-- 中文摘要：Machine learning approaches to spatiotemporal physical systems have primarily focused on next-frame prediction, with the goal of learning an accurate emulator for the system's evolution in time. However, these emulators are computationally expensive to train and are subject to performance pitfalls, such as compounding errors during autoregressive rollout. In this work, we take a different perspective and look at scientific tasks further downstre…
+- 标签：`表征学习` `潜在空间` `世界模型` `预训练`
+- 中文摘要：【LLM 暂不可用，先保留英文摘要要点】Machine learning approaches to spatiotemporal physical systems have primarily focused on next-frame prediction, with the goal of learning an accurate emulator for the system's evolution in time. However, these emulators…
 - 链接：[DOI](https://doi.org/10.48550/arXiv.2603.13227) | [arXiv](https://arxiv.org/abs/2603.13227v1) | [PDF](https://arxiv.org/pdf/2603.13227v1)
 
 ### 2. Separable neural architectures as a primitive for unified predictive and generative intelligence
-> 应用信号: system, navigation, autonomous；创新信号: unified；领域匹配: representation, representations, embedding
+> Separable neural architectures as a primitive for unified predictive and generative i…
 - 作者：Reza T. Batley, Apurba Sarker, Rajib Mostakim, Andrew Klichine 等另外1人
-- 标签：—
-- 中文摘要：Intelligent systems across physics, language and perception often exhibit factorisable structure, yet are typically modelled by monolithic neural architectures that do not explicitly exploit this structure. The separable neural architecture (SNA) addresses this by formalising a representational class that unifies additive, quadratic and tensor-decomposed neural models. By constraining interaction order and tensor rank, SNAs impose a structural i…
+- 标签：`表征学习` `潜在空间` `世界模型` `预训练`
+- 中文摘要：【LLM 暂不可用，先保留英文摘要要点】Intelligent systems across physics, language and perception often exhibit factorisable structure, yet are typically modelled by monolithic neural architectures that do not explicitly exploit this structure. The separabl…
 - 链接：[DOI](https://doi.org/10.48550/arXiv.2603.12244) | [arXiv](https://arxiv.org/abs/2603.12244v1) | [PDF](https://arxiv.org/pdf/2603.12244v1)
 
 ### 3. VIRD: View-Invariant Representation through Dual-Axis Transformation for Cross-View Pose Estimation
-> 应用信号: robot, robotic, dataset；创新信号: novel, first；领域匹配: representation, representations, feature
+> VIRD，采用表征学习解决机器人操作，实现首次提出
 - 作者：Juhye Park, Wooju Lee, Dasol Hong, Changki Sung 等另外3人
-- 标签：—
-- 中文摘要：Accurate global localization is crucial for autonomous driving and robotics, but GNSS-based approaches often degrade due to occlusion and multipath effects. As an emerging alternative, cross-view pose estimation predicts the 3-DoF camera pose corresponding to a ground-view image with respect to a geo-referenced satellite image. However, existing methods struggle to bridge the significant viewpoint gap between the ground and satellite views mainl…
+- 标签：`表征学习` `潜在空间` `世界模型` `预训练`
+- 中文摘要：【LLM 暂不可用，先保留英文摘要要点】Accurate global localization is crucial for autonomous driving and robotics, but GNSS-based approaches often degrade due to occlusion and multipath effects. As an emerging alternative, cross-view pose estimation predict…
 - 链接：[DOI](https://doi.org/10.48550/arXiv.2603.12918) | [arXiv](https://arxiv.org/abs/2603.12918v1) | [PDF](https://arxiv.org/pdf/2603.12918v1)
 
 ## 强化学习（2 篇）
 
 ### 1. Efficient Real-World Autonomous Racing via Attenuated Residual Policy Optimization
-> 应用信号: real-world, deployment, robot；创新信号: zero-shot；领域匹配: reinforcement learning, policy optimization, control
+> Efficient Real-World Autonomous Racing via Attenuated Residual Policy Optimization，采用…
 - 作者：Raphael Trumpp, Denis Hoornaert, Mirco Theile, Marco Caccamo
-- 标签：—
-- 中文摘要：Residual policy learning (RPL), in which a learned policy refines a static base policy using deep reinforcement learning (DRL), has shown strong performance across various robotic applications. Its effectiveness is particularly evident in autonomous racing, a domain that serves as a challenging benchmark for real-world DRL. However, deploying RPL-based controllers introduces system complexity and increases inference latency. We address this by i…
+- 标签：`强化学习` `策略优化` `奖励设计` `离线 RL`
+- 中文摘要：【LLM 暂不可用，先保留英文摘要要点】Residual policy learning (RPL), in which a learned policy refines a static base policy using deep reinforcement learning (DRL), has shown strong performance across various robotic applications. Its effectiveness is part…
 - 链接：[DOI](https://doi.org/10.48550/arXiv.2603.12960) | [arXiv](https://arxiv.org/abs/2603.12960v1) | [PDF](https://arxiv.org/pdf/2603.12960v1)
 
 ### 2. Beyond Imitation: Reinforcement Learning Fine-Tuning for Adaptive Diffusion Navigation Policies
-> 应用信号: deployed, robot, dataset；创新信号: diffusion, zero-shot；领域匹配: reinforcement learning, policy optimization, imitation learning
+> Beyond Imitation，采用扩散模型解决机器人操作，实现零样本泛化
 - 作者：Junhe Sheng, Ruofei Bai, Kuan Xu, Ruimeng Liu 等另外4人
-- 标签：—
-- 中文摘要：Diffusion-based robot navigation policies trained on large-scale imitation learning datasets, can generate multi-modal trajectories directly from the robot's visual observations, bypassing the traditional localization-mapping-planning pipeline and achieving strong zero-shot generalization. However, their performance remains constrained by the coverage of offline datasets, and when deployed in unseen settings, distribution shift often leads to ac…
+- 标签：`强化学习` `策略优化` `奖励设计` `离线 RL`
+- 中文摘要：【LLM 暂不可用，先保留英文摘要要点】Diffusion-based robot navigation policies trained on large-scale imitation learning datasets, can generate multi-modal trajectories directly from the robot's visual observations, bypassing the traditional localization-m…
 - 链接：[DOI](https://doi.org/10.48550/arXiv.2603.12868) | [arXiv](https://arxiv.org/abs/2603.12868v1) | [PDF](https://arxiv.org/pdf/2603.12868v1)