---
title: "Cortex: A Bidirectionally Aligned Embodied Agent Framework for Long-horizon Manipulation"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.05377v1"
permalink: "https://feed.astralor.com/items/itm_7a3ae66350397ad980c4d73c"
published: "2026-07-06T17:55:05.000Z"
score: 81
channel: "ai"
tags: ["Cortex","Vision-Language-Action","Embodied AI","Long-horizon Manipulation","zero-shot generalization"]
---

# Cortex: A Bidirectionally Aligned Embodied Agent Framework for Long-horizon Manipulation

## AI commentary

> What stands out to me in Cortex is how it bridges the gap between high-level semantic planning and low-level motion control with a clever set of skill primitives and tractability principles. This isn't just a paper; it reads like a deployable system design—especially the zero-shot chemistry experiment demo, which hints at the potential for generalist robot manipulation. It's a must-read for those working on robot learning, VLA models, and industrial automation.

## At a glance

Cortex is a bidirectionally aligned embodied agent framework that bridges high-level VLM planning semantics and low-level VLA execution kinematics by standardizing 32 canonical skill primitives and injecting tractability principles. It leverages over 4k hours of open-source video data and 30 hours of simulation data for automated annotation, fine-tunes with event-balanced sampling, and outperforms monolithic baselines by 3.1% on Libero-long and 4.1% on RoboTwin, enabling zero-shot completion of unseen real-world long-horizon tasks like multi-stage chemistry experiments.

## Key points

- Cortex unifies 32 canonical manipulation skill primitives and injects tractability principles, resolving the semantic gap between high-level planning and low-level execution.
- Automatically annotates over 4k hours of open-source video and generates 30 hours of simulation data for large-scale training.
- Event-balanced sampling effectively handles planning ambiguity during subtask transitions, improving long-horizon robustness.
- Zero-shot generalization to novel real-world tasks (e.g., chemistry experiments) demonstrates versatility beyond VLA fine-tuning alone.

## Full text

While recent Vision-Language-Action (VLA) models show promise toward generalist manipulation policies, they struggle with long-horizon tasks due to their Markovian nature-relying solely on current observations. Hierarchical dual-system methods address this but suffer from a gap between high-level planning semantics and low-level execution kinematics. We introduce Cortex, a bidirectionally aligned embodied agent framework with a customized planning interface that conveys executable and tractable subtask plans from high-level VLM to low-level VLA. Specifically, we standardize manipulation subtasks into 32 canonical skill primitives and inject tractability principles, such as representative object attributes and improved trajectory reachability, into the data generation pipeline. This enables automatic annotation of over 4k hours of open-source video data and generation of 30 hours of simulation data. We further devise an event-balanced sampling strategy to construct training data for fine-tuning the framework to better handle planning ambiguity during subtask transitions, enhanced by carefully designed harness engineering from task contexts to skill constraints during inference. Both open-loop VLM and closed-loop system evaluations demonstrate Cortex's efficacy, e.g., it outperforms monolithic baselines by 3.1% on Libero-long and 4.1% on RoboTwin. Notably, Cortex's generalist VLM enables zero-shot completion of unseen real-world long-horizon tasks, such as multi-stage chemistry experiments, by simply combining with a fine-tuned VLA-a capability infeasible through VLA fine-tuning alone.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_7a3ae66350397ad980c4d73c