DAILY INTELLIGENCE · AI APP + VLA
研究日報研究日报
AI App 精選與 VLA 論文評級的每日合輯AI App 精选与 VLA 论文评级的每日合辑。 下方匯總近 7 天的信號總覽，歸檔索引可按領域過濾下方汇总近 7 天的信号总览，归档索引可按领域过滤。
近 7 日匯總近 7 日汇总 10-19 → 10-26
56 AI精選AI精选
185 VLA論文VLA论文
⚡ 0 突破
🔧 31 工具/技術工具/技术
📖 154 背景/觀點背景/观点
近 3 天內容近 3 天内容 10-24 → 10-26
2026-10-26 最新 47 條条
OpenAI Decisions API 正式 GA：基于 GPT-6 Luna，决策延迟压到 100-180ms、比走 Responses API 快最多 10x AI Kimi K3.1 Pro 限流事故致歉：全体订阅会员补偿月度额度 25%，根因是不必要的子 Agent 调用 AI JetBrains 开源 ThinkRail：围绕 pi coding agent 重建的 IDE，git worktree 隔离工作区 + 自生长 skills AI 五实例共享文件系统自编排：Qwen3.8-27B 从 66.8% 拉到 92.4% LiveCodeBench-Hard（单 GPU 可跑） AI HN 热议：OpenAI 与 Anthropic 数月内把 LLM 价格砍到最高 -80%，降价战逻辑拆解 AI Google Gemini 4 Argon 进入 Antigravity：Alphabet 内部 10 万+ 用户实测，agentic 代码 +35%、提交后回滚 -30% AI MakeCindy/cindy：开箱即用的开源 AI Agent（TypeScript，4.8k⭐，昨日仍在发版） AI PrefO2O: Zero-Shot Preference-Based Reinforcement Learning via Offline-to-Online Latent Preference Transfer VLA WorldMirror: Agentic Real-to-Sim Generation of Interactive Worlds from In-the-Wild Robot Demonstrations VLA FORCA: A Force-Aware Coding Agent for Contact-Rich Robot Manipulation VLA EgoForesight: Effectively Scaling Policy Learning with Noisy Egocentric Actions VLA Full-Weight VLA Fine-Tuning Without an fp32 Master Copy Silently Erases Most Updates VLA DiscoDemo: Discovering Efficient and Diverse Robot Demonstrations for Imitation Learning VLA TTT in VLA Models: Exploring Training with Human Demonstrations VLA Coupling Execution Horizon and Sampling Depth in Flow-Matching VLA Policies VLA Do VLAs Know They Will Fail? Interpretable Success and Failure Patterns in Physical AI Systems VLA RoboCode: Learning Reusable Action Primitives for Robot Manipulation VLA From Tabletop to Walking Bot: Deploying a VLA Model on a Legged Manipulator Robot VLA MAST-VLA: Skill-Conditioned Adaptation with Trajectory-Affordance Priors for Multi-Skill Robot Manipulation VLA DynaSlots: Slot-Centric 3D Dynamics Pretraining for Robot Manipulation VLA Predictive Long-Horizon Progress Modeling with Action-Aware Quantification VLA When Human Corrections Travel: Safety in Federated VLA Adaptation VLA If at First You Don't Succeed, Backtrack: Lightweight Recovery for Frozen VLAs VLA DexPort: A Hand-Held Skill Interface for Dexterous Hand Data Collection VLA Retiming Robot Demonstrations with a Learned Speed Prior VLA RobustRLlib: A Unified Library and Benchmark for Robust Reinforcement Learning Algorithms VLA Emergent Neural Automaton Policies: Learning Symbolic Structure from Visuomotor Trajectories VLA 3D Generation for Embodied AI and Robotic Simulation: A Survey VLA Language as a Sensor: Calibrated Spatial Belief Estimation in 3D Scenes from Natural Language VLA Inference-Time Robot Behavior Steering through Physically-Aware Reconfiguration of Task-Structure VLA Semantic Anchoring for Robotic Action Representations VLA Unified Condition-Action Modeling for Accurate One-Step Action Generation VLA ContactBench 1.0: A Benchmark for Everyday Contact-Rich Robotic Manipulation VLA DepState-JEPA: Learning Depth-Grounded State Representation for Visual Perception and Manipulation VLA Training in Deployment: Human-in-the-loop Online Post-Training for Robot Manipulation VLA Sensorimotor Alignment: A Platonic Proxy for Embodied Policy Success VLA When to Stop Penalizing Actions? Martingale GRPO for VLA Reinforcement Learning VLA HGWM: Hierarchical Graph-Guided World Model for Zero-Shot Object Navigation via Graph Matching VLA Regularised RL improves sim-to-real transfer VLA WAM-TTT: Steering World-Action Models by Watching Human Play at Test Time VLA Do Robotic World Models Really Follow Actions? Diagnosing and Aligning Action-Conditioned Generation for Policy Learning VLA $\Psi$-0.6: Adaptive Streams of Thought in a Unified Vision-Language World Model VLA Learning from Next-Frame Prediction: Towards Unified and Scalable Visual World Modeling VLA UniWorld: A Unified World Model for Real-Time Visual World Simulation VLA UniAction: Unified Action Conditioning for World Models VLA VisFlow: Interpretable Flow Matching for Visibility Reconstruction, Characterization, and Recognition VLA Trajectory Matching: Discrete Flow Matching with Structured Actions VLA
2026-10-25 昨日 9 條条
Kimi K3.1 Pro 上线：AA 智能指数 55 分，距前沿只差 3 分 AI Thariq Shihipar 离职 Anthropic：称大模型智力已过剩 AI NVIDIA OpenShell：内核级沙箱给 agent 舰队上政策 AI Moonshot 开放前删减网络攻击训练数据 AI AWS Agent Toolkit：官方 MCP servers + skills + plugins AI Nutlope/hallmark：反 AI slop 的前端设计 skill AI TRAE 1024 大赏：开发者最爱的 10 个 MCP Server AI OpenAI 再表态前沿模型短期不开源 AI Show HN：CVE-2022-22706 拿到 TCL Android TV root AI
2026-10-24 2日前 11 條条
Kimi K3.1 Pro 发布：3T 级开源权重、100 万上下文，权重 11/04 前放出 AI Anthropic 详解 Claude Code auto mode：分类器代替 93% 的人工批准 AI Kimi K3.1 Pro Highspeed 同步上线：同等智能、输出 2-3 倍速 AI OpenAI 认为现有安全措施足以支持重启前沿强化学习训练 AI DeepSeek 完成约 1000 亿元新一轮融资，宁德时代拟跟投 100 亿 AI 科大讯飞 1024 开发者节：星火 X2.6、SparkAudio1.0 与 Agent OS AI Simon Willison：一句 prompt 让 Claude 用 canvas 几秒重绘照片成油画 AI 用 Claude 给小米电视去广告（社区热帖） AI vercel/eve — filesystem-first agent 框架 AI PrimeIntellect-ai/prime-agent — 自改进 RLM coding agent AI DRILL: Training World Models to Improve Policies, Not Predict Pixels VLA
歸檔索引归档索引
66 期— 2026 年 10 月 —
AI 最新
7 篇
VLA 最新
40 篇
AI 昨日
9 篇
AI 2日前
10 篇
VLA 2日前
1 篇
AI
10 篇
VLA
28 篇
AI
6 篇
VLA
28 篇
AI
7 篇
VLA
29 篇
AI
7 篇
VLA
29 篇
AI
9 篇
VLA
30 篇
AI
10 篇
AI
9 篇
VLA
22 篇
AI
7 篇
VLA
30 篇
VLA
29 篇
AI
6 篇
AI
8 篇
VLA
29 篇
VLA
1 篇
VLA
23 篇
VLA
54 篇
VLA
48 篇
VLA
49 篇
AI
8 篇
VLA
29 篇
AI
7 篇
VLA
1 篇
AI
7 篇
VLA
1 篇
AI
10 篇
VLA
29 篇
AI
9 篇
VLA
27 篇
AI
8 篇
VLA
30 篇
— 2026 年 9 月 —
AI
7 篇
VLA
29 篇
AI
8 篇
VLA
28 篇
AI
7 篇
AI
7 篇
AI
9 篇
VLA
30 篇
VLA
28 篇
VLA
29 篇
VLA
29 篇
VLA
30 篇
VLA
30 篇
VLA
28 篇
VLA
31 篇
VLA
27 篇
VLA
27 篇
VLA
1 篇
VLA
28 篇
VLA
27 篇
VLA
36 篇
VLA
9 篇
VLA
35 篇
VLA
10 篇
VLA
10 篇