GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness For Variational Automation Tasks

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: 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
• Venue: arXiv: cs.RO
• Published: 2026-07-06
• Collected: 2026-07-07
• URL: http://arxiv.org/abs/2607.05369v1
• PDF: https://arxiv.org/pdf/2607.05369v1
• DOI: UNKNOWN

Plain-Language Brief

The paper introduces a new multi-agent coding harness called Graph-as-Policy (GaP) for improving the reliability of robots in commercial and industrial applications, specifically focusing on variational automation tasks. GaP generates directed computation graphs from a Modular Open Robot Skill Library (MORSL), then simulates and refines these graphs to enhance success rates and throughput. The results show that GaP outperforms baselines on new open VA task benchmarks, both in simulation and real-world scenarios.

What the Document Says

• The focus is on variational automation tasks for improving robot reliability in commercial and industrial applications.
• GaP is a multi-agent coding harness that generates directed computation graphs from MORSL.
• GaP simulates task instances with different graphs to iteratively refine the graph structure and parameters.
• The evaluation involves 8 new open VA task benchmarks, both in simulation and real-world scenarios.
• GaP outperforms baselines according to the success rates reported.

Key Concepts Mentioned

• robotics
• automation
• graph-as-policy
• multi-agent systems
• variational automation
• task benchmarks
• MORSL

Claims Made by the Document

• GaP can improve the reliability of robots in commercial and industrial applications.
• GaP outperforms baselines on new open VA task benchmarks.

Evidence Used

• The focus, introduction, and evaluation sections of the paper

Methods / Approach

• Generating directed computation graphs from MORSL and simulating task instances to refine graph structure and parameters

Limitations Stated by the Authors

The paper does not provide specific details about the baselines for comparison.

Limitations Noted by the Agent

Agent interpretation — not from the source.

The paper does not specify the criteria used to determine success rates or throughput.

Open Questions

• What are the specific baselines used in the evaluation?
• How exactly were success rates and throughput measured?

Notes for Future Q&A

May include interpretation. Labeled where it does.

• It would be beneficial to know more about the baselines mentioned, as they are crucial for understanding the context of the paper's claims.

Raw Abstract / Excerpt

For robots to work reliably in commercial and industrial applications, can recent advances in agentic coding systems combine interpretable robot programming with the open-world adaptability of model-free policies? We focus on 'Variational Automation' (VA), a class of tasks that have larger variations in object geometry and pose than fixed automation. Model-free policies often struggle to close the reliability gap for VA tasks, which must be executed persistently and reliably in commercial and industrial applications. Motivated by prior work on Task and Motion Planning (TAMP) and the Robot Operating System (ROS), we introduce Graph-as-Policy (GaP), a multi-agent coding harness that generates directed computation graphs with perception, planning, and control nodes from a Modular Open Robot Skill Library (MORSL). GaP then generates an internal simulation environment to rehearse task instances with different graphs in parallel to iteratively refine the graph structure and parameters to improve success rates and throughput. Evaluation with 8 new open VA task benchmarks, 4 in-simulation and 4 in real-world, suggests that GaP can achieve success rates that significantly outperform baselines. Details, code, and data can be found online: https://graph-robots.github.io/gap