EvoPolicyGym: Evaluating Autonomous Policy Evolution in Interactive Environments

Hugging Face Daily Papers Papers

Summary

EvoPolicyGym is a benchmark for evaluating how autonomous agents iteratively improve policies through feedback in interactive environments, with GPT-5.5 achieving top performance.

Autonomous agents are increasingly expected to improve executable policies through feedback, yet existing evaluations often collapse this process into a final score or confound it with open-ended software-engineering progress. We introduce Autonomous Policy Evolution, a controlled evaluation setting in which a harness-model agent repeatedly edits an executable policy system under a fixed interaction budget. We instantiate this setting in EvoPolicyGym, a benchmark built from compact interactive RL environments that evaluates how agents iteratively improve explored policies. On the EvoPolicyGym suite, GPT-5.5 achieves the strongest aggregate rank score and top-two performance on all 16 environments. Beyond leaderboard results, EvoPolicyGym also provides trajectory-level diagnostics that distinguish how agents allocate budget, convert feedback into parametric tuning. These analyses show that strong autonomous policy evolution depends not only on isolated task wins, but on discovering task-appropriate mechanisms and refining policies under bounded feedback.
Original Article
View Cached Full Text

Cached at: 07/03/26, 07:53 AM

Paper page - EvoPolicyGym: Evaluating Autonomous Policy Evolution in Interactive Environments

Source: https://huggingface.co/papers/2607.02440 Published on Jul 2

#3 Paper of the day Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

Autonomous agents evaluate policy improvement through iterative editing within fixed budgets, revealing that successful policy evolution requires both task-specific mechanisms and feedback-constrained refinement.

Autonomous agentsare increasingly expected to improveexecutable policiesthrough feedback, yet existing evaluations often collapse this process into a final score or confound it with open-ended software-engineering progress. We introduce AutonomousPolicy Evolution, a controlled evaluation setting in which a harness-model agent repeatedly edits an executable policy system under a fixed interaction budget. We instantiate this setting in EvoPolicyGym, a benchmark built from compact interactive RL environments that evaluates how agents iteratively improve explored policies. On the EvoPolicyGym suite, GPT-5.5 achieves the strongest aggregate rank score and top-two performance on all 16 environments. Beyond leaderboard results, EvoPolicyGym also providestrajectory-level diagnosticsthat distinguish how agents allocate budget, convert feedback intoparametric tuning. These analyses show that strong autonomouspolicy evolutiondepends not only on isolated task wins, but on discoveringtask-appropriate mechanismsand refining policies under bounded feedback.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2607\.02440

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.02440 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.02440 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.02440 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Evolved Policy Gradients

OpenAI Blog

OpenAI introduces Evolved Policy Gradients (EPG), a meta-learning approach that learns loss functions through evolution rather than learning policies directly, enabling RL agents to generalize better across tasks by leveraging prior experience similar to how humans transfer skills.

EvoCUA-1.5: Online Reinforcement Learning for Multi-turn Computer-Use Agents

arXiv cs.AI

EvoCUA-1.5 introduces an online reinforcement learning framework for multi-turn computer-use agents, achieving a 63.2% success rate on OSWorld-Verified and outperforming comparable open-weight models up to 35B parameters through step-level policy optimization and dynamic curriculum learning.