Tag
提出了Orchestra，一种将任务规划与执行管理同动作生成分离的机器人智能体框架，用于执行多步多模态指令的长时程操作，并发布了基准RMBench-Compose。Orchestra在多个基准和真实世界多步任务上显著超越Mem-0和HarnessVLA等基线。
The paper presents TopoPlanner, a topology-consistent planning framework that lifts tool dependency graphs into cellular workflow complexes to guide LLM tool planning, enabling handling of verification-correction loops, branch merging, and reusable intermediate states beyond sequential or DAG-like workflows, with consistent gains over baselines on four benchmarks.
DAGent (NeurIPS 2026) introduces an Evaluate-then-Grow DAG-based multi-agent framework where an Orchestrator expands the task graph batch-by-batch based on node confidence, along with DAGRPO, a topology-conditioned RL algorithm that outperforms GRPO on deep research benchmarks.
A fully differentiable neuro-soft-symbolic framework integrates visual perception and task planning in a single computational graph, achieving higher success rates on benchmarks like Blocksworld and PlanBench with reduced computation.
X-Planner introduces an event-structured planning front-end for embodied AI, improving task planning in long-horizon manipulation through supervised data and a VLM backbone with discrete and latent interfaces.
This paper delineates the capability boundaries of multi-agent collaboration in LLM-based systems, showing benefits only in specific task structures like long-horizon tasks with sparse dependencies, and proposes SAIGE, a dynamic graph-based mechanism for efficient collaboration.
Ordewell is a CLI tool that turns a single goal into an ordered plan of coding-agent tasks, each with customizable runners, models, and modes, then executes and verifies the plan.
This article introduces three third-party skills for Codex, including browser automation, file planning, and coding guidelines, to improve the performance and reliability of AI agents.
Meta AI's new task-planning features raise concerns about conflicts of interest given Meta's advertising and commerce operations, prompting discussion of disclosure policies.
Researchers from Tsinghua and South China University of Technology introduced Atomic Task Graph (ATG), a framework that enables 7B-8B open-source models to surpass GPT-4 on complex agent benchmarks without fine-tuning, by using directed graph-based planning and internal simulation to drastically reduce hallucination rates.
MultiUAV-Plat is a lightweight simulation platform and benchmark for evaluating LLM agents on multi-UAV collaborative task planning tasks. The paper also proposes Agent4Drone, a task-specific LLM agent framework that significantly outperforms baselines.
UI-KOBE proposes a framework that enhances lightweight mobile GUI agents by constructing and leveraging app-specific knowledge graphs to improve task planning and execution efficiency.
The video demonstrates the ability of multiple robots under the Gemini Robotics 2 system to collaborate on cleaning and organizing tasks through natural language instructions, showcasing task allocation and coordination between robots.