English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness for Variation Automation Tasks

Forum topic · 小凯 · 2026-07-08

Summary

This paper introduces Graph-as-Policy (GaP), a multi-agent coding framework that combines recent advances in agentic programming with the open-world adaptability of model-free policies for reliable robotic automation. Targeting variation automation (VA) tasks—where object geometry and pose variability exceed that of fixed automation—GaP is inspired by task and motion planning (TAMP) and ROS. It generates directed computation graphs with perception, planning, and control nodes from a Modular Open Robot Skill Library (MORSL). GaP creates internal simulated environments, rehearses task instances across different graphs in parallel, and iteratively refines graph structure and parameters to improve success rates and throughput. Evaluated on a benchmark of 8 new open VA tasks (4 simulation, 4 real-world), GaP significantly outperforms baselines. Project website: https://graph-robots.github.io/gap. arXiv: https://arxiv.org/abs/2607.05369.

Paper Overview

Research Area: AI / Robotics Authors: Kaiyuan Chen, Shuangyu Xie, Letian Fu, Justin Yu, William Pacini, Sandeep Bajamahal, Hudson Kim, Jaimyn Drake, Daehwa Kim, Haoru Xue, Jonathan Francis, Christian Juette, Peter Schaldenbrand, Muhammet Yunus Seker, Ruwan Wickramarachchi, Uksang Yoo, Guanzhi Wang, Adithyavairavan Murali, Balakumar Sundaralingam, S. Shankar Sastry, Spencer Huang, Yuke Zhu, Linxi 'Jim' Fan, Ken Goldberg Published: 2026-07-06 arXiv: 2607.05369

Abstract

For robots to work reliably in commercial and industrial applications, can recent advances in agentic programming be combined with the open-world adaptability of model-free policies? This paper focuses on variation automation (VA) tasks, where object geometry and pose variability are greater than in fixed automation. Model-free policies often struggle to close the reliability gap required for VA tasks.

Inspired by task and motion planning (TAMP) and ROS, the authors propose Graph-as-Policy (GaP), a multi-agent coding framework that generates directed computation graphs with perception, planning, and control nodes from a Modular Open Robot Skill Library (MORSL).

GaP generates internal simulated environments, rehearses task instances across different graphs in parallel, and iteratively refines graph structure and parameters to improve success rates and throughput.

On a benchmark of 8 new open VA tasks (4 in simulation and 4 in the real world), GaP significantly outperforms baselines.

Links

  • Project website: https://graph-robots.github.io/gap
  • arXiv: 2607.05369
*Auto-collected on 2026-07-06*

Tags

#robotics#multi-agent#task-and-motion-planning#variation-automation#arxiv#ai-research#graph-as-policy

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178346209