planning

Tag

Cards List
#planning

Capability-Grounded Agentic Planning

arXiv cs.AI ↗ · 2026-11-02 Cached

The paper presents CGAP, a neuro-symbolic framework that compiles tool catalogs into a PDDL model and uses a classical planner with LLM-based goal parsing to guarantee executable multi-tool plans, outperforming standard tool-calling agents by up to 28.9 points on four benchmarks.

0 favorites 0 likes
#planning

@KirkDBorne: New survey report >> Agentic Reasoning for Large Language Models: https://arxiv.org/abs/2601.12538 [135-page PDF]

X AI KOLs Timeline ↗ · 2026-11-01 Cached

A 135-page survey paper organizing agentic reasoning for large language models along three layers—foundational, self-evolving, and collective multi-agent reasoning—spanning in-context and post-training optimization with real-world applications and open challenges.

0 favorites 0 likes
#planning

Second Thoughts: Exploring and Planning through Comparative Imagination in World Models

arXiv cs.LG ↗ · 2026-10-22 Cached

The paper introduces SDI (Sibling Disagreement in Imagination), a method that unifies exploration and planning in world-model-based RL by using the actor's own rollout variability as a shared reference to decide when a planned action should replace the learned policy, outperforming baselines on Meta-World, Adroit, and DM-Control.

0 favorites 0 likes
#planning

M-Plan: Comprehensive Planning For LLM Agents with Memory Arrangement

arXiv cs.AI ↗ · 2026-10-21 Cached

M-Plan is a closed-loop multi-agent system framework that integrates long-term and short-term memory with global-local and past-current guidance to improve LLM agents' planning on long-horizon multi-turn tasks, outperforming baselines on ALFWorld, Webshop, and PDDL.

0 favorites 0 likes
#planning

Latent Projection for Latent Planning

arXiv cs.LG ↗ · 2026-10-21 Cached

This paper proposes a post-hoc latent projection method for latent world model planning that prunes noisy latent directions by fitting the state dynamics matrix and restricting the ℓ2 planning cost to the least-uncertain subspace, reporting up to +54.7 points success-rate gains on 2D navigation tasks like TwoRoom and PointMaze.

0 favorites 0 likes
#planning

CausGround: Causal Interaction Grounding for Adaptive Embodied Planning

Hugging Face Daily Papers ↗ · 2026-10-21 Cached

CausGround 提出一种非参数化的因果交互 grounding 框架,从交互经验中提炼可复用的动作前置条件与效果,无需重训模型即可提升 MLLM 在 VirtualHome、RoboCasa 等基准上的具身规划成功率(如 24.9% 提升至 53.3%)。

0 favorites 0 likes
#planning

World Models with Configurable Perception

Hugging Face Daily Papers ↗ · 2026-10-20 Cached

ConfigWM introduces configurable perception for world models, using task descriptions to steer visual embeddings into task-specific state spaces that filter out irrelevant distractions, achieving 22.3% better planning success over the strongest baseline in noisy environments.

0 favorites 0 likes
#planning

Testing the Planning Claims of LLM-Based Agents against Formal Semantics

arXiv cs.AI ↗ · 2026-10-19 Cached

该论文通过隐藏的PDDL形式语义对LLM智能体的规划能力进行了严格验证,发现最佳配置离线可解78%的任务但在线仅40%,且复杂提示策略在无明显收益的情况下多消耗5至22倍token,质疑了现有LLM智能体规划能力的声明。

0 favorites 0 likes
#planning

From Events to Reasoning: Hierarchical Rule Composition over Explicit States

arXiv cs.AI ↗ · 2026-10-19 Cached

论文提出事件推理系统(ERS),将推理状态显式表示为事件图,通过低层可复用规则与可递归组合的高层控制规则构建可解释、可修订的推理路径,在结构化事件图上实现长依赖链、零无关操作与高效的动态子目标控制。

0 favorites 0 likes
#planning

AnyGoal: On World Model Generalization Beyond Expert Rollouts

Hugging Face Daily Papers ↗ · 2026-10-19 Cached

研究人员提出 AnyGoal,一个基于目标偏移任务变体的世界模型基准与开源库,用于评估世界模型能否将熟悉的状态与动作重新组合成有效动力学。实验发现现有世界模型家族在这些重组场景下普遍失效,揭示了动力学泛化与规划性能之间的关键差距。

0 favorites 0 likes
#planning

SHRED: Self-Play for Heterogeneous Road-Users without Expert Demonstrations

arXiv cs.LG ↗ · 2026-10-16 Cached

SHRED 是一个通过大规模异构自博弈(车辆、行人、骑行者)从随机初始化联合训练的框架,无需人类驾驶演示或日志轨迹回放,即可学习真实且交互性强的交通行为。在 Waymo Open Sim Agents Challenge 和 InterPlan 基准上,SHRED 优于依赖演示的自博弈方法和基线规划器。

0 favorites 0 likes
#planning

CarryOn: A Dual-Loop Harness for Personalized and Reliable Robot Assistance

Hugging Face Daily Papers ↗ · 2026-10-15 Cached

CarryOn 提出一个双循环框架,在冻结的视觉-语言-动作(VLA)策略上加入持久化个性化规划与执行监督,显著降低安全违规率并提升故障恢复成功率。

0 favorites 0 likes
#planning

@mattpocockuk: Debating adding a GRILL.md file to customize how /grill-me works A few benefits: 1. /grill-with-docs gets removed, you …

X AI KOLs Timeline ↗ · 2026-10-14 Cached

Matt Pocock 讨论为 /grill-me 技能添加 GRILL.md 配置文件,让用户能自定义规划阶段的审查重点(如安全性、模块策略)以及逐项或整体提问的方式。

0 favorites 0 likes
#planning

@rohanpaul_ai: A stronger robot doesn't always require a stronger foundation model. Sometimes, it just needs a better way to act. Rece…

X AI KOLs Timeline ↗ · 2026-10-10 Cached

SenseNova-RoboRSI 研究展示了通过智能体系统设计(而非更换更强基础模型)来提升机器人性能的方法,在模型权重不变的情况下,RoboDojo 平均分从 28.97 提升至 56.83,并结合递归自我改进(RSI)循环优化 agent harness 配置。

0 favorites 0 likes
#planning

PANINI-DR: Controlled Deep Research through Generative Semantic Workspaces

Hugging Face Daily Papers ↗ · 2026-10-10 Cached

PANINI-DR提出了一种多智能体深度研究框架,通过生成式语义工作区分离协调、研究与验证职责,在BrowseComp-Plus等三个基准上取得最佳平均性能,BrowseComp-Plus准确率达73.3%,领先次优系统16.0个百分点。

0 favorites 0 likes
#planning

TaReD: Tool-Aware Recursive Decomposition for Long-Horizon Tasks

arXiv cs.AI ↗ · 2026-10-09 Cached

TaReD 提出一种工具感知的递归分解方法,通过按功能关系将工具组织成能力层级,让智能体按需发现工具并将复杂长时程任务递归分解为子任务树,实验显示端到端任务成功率最高提升 40 个百分点。

0 favorites 0 likes
#planning

Learning How to Search for Plans with Exponentially Less Space

arXiv cs.AI ↗ · 2026-10-09 Cached

该论文提出通过学习索引策略(indexical policies)形式的搜索控制,使规划搜索在多项式空间内完成,代价是时间仅随选择深度指数增长;结合语言模型的反例引导学习循环,在IPC 2023学习赛道等基准上解决了1,709/1,890个测试任务,优于LAMA、BFWS和Levitron。

0 favorites 0 likes
#planning

Plan-and-Patch: Diffusion Language Models for Agentic Planning

arXiv cs.AI ↗ · 2026-10-09 Cached

提出 Plan-and-Patch 框架,利用扩散语言模型生成结构化计划并通过局部修补(保留前后缀、重新生成受影响区域)实现高效计划修复;在 Natural Plan 上无任务训练时修复成功率约为自回归模型的两倍,并将计划生成延迟降低 39-46%。

0 favorites 0 likes
#planning

LeWAM: End-to-End World and Action Modeling with JEPAs

Hugging Face Daily Papers ↗ · 2026-10-08 Cached

LeWAM 提出一种端到端的 JEPA 控制框架,通过联合训练下一隐状态预测与动作流匹配目标,并对隐状态施加各向异性高斯约束,使单一模型同时支持直接策略执行与高效规划,在接触丰富任务和长程任务上将成功率从 28.6% 提升至 89.7% 和从 10.9% 提升至 46.2%,规划速度提升 32.3 倍。

0 favorites 0 likes
#planning

Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks

Hugging Face Daily Papers ↗ · 2026-10-08 Cached

Memento 3 让冻结的 LLM 智能体通过自然语言规则簿与代码编译循环,实现基于模型的递归自我改进,在 ARC-AGI-3 上清空全部 25 个公开游戏关卡并达到 100.0 的平均 RHAE。

0 favorites 0 likes
Next →
← Back to home

Submit Feedback