Verifiable Hidden Dynamics Play: Generating Agentic RL Environments from Solved Mechanisms

Hugging Face Daily Papers Papers

Summary

This paper introduces VHD-Play, a pipeline that generates diverse agentic reinforcement learning environments by first solving mathematical models, significantly improving training for language-model agents like Qwen3.6-35B-A3B at low cost and extending to external benchmarks.

Language-model agents increasingly face long-horizon tasks with evolving state, interdependent decisions, and delayed outcomes. Scaling their training requires diverse agentic environments, dependable outcome signals, and low extension cost. Existing generation pipelines commonly construct an environment before defining its outcome rule or annotating its trajectories, leaving dynamics and evaluation to be aligned post hoc. VHD-Play reverses this dependency by sampling and solving a mathematical model before a corpus-grounded setter renders its decision process as stateful tools. The executable dynamics and trajectory-scoring reference are inherited from the same solved model. The pipeline produces 3,300 diverse agentic environments at a cost of a few cents each. Training Qwen3.6-35B-A3B on three families raises its mean agentic score from 0.204 to 0.815 in a five-family diagnostic. Gains also appear on held-out instances from all three training families and eight unseen mechanism families, then extend beyond the generated substrate to external benchmarks for general function calling, travel planning, and 365-day e-commerce. On E-Commerce Bench, the trained checkpoint completes every run without bankruptcy and exceeds Qwen3.7-Max. We compare written-out problems with stateful versions that reveal or hide their parameters. The comparison shows that most of the learnable gap lies in stateful interaction rather than underlying problem solving. A frozen 35B setter realizes larger environments, and scale-matched training retains gains as mechanism size and horizon grow, indicating the potential for an evolving training substrate.
Original Article
View Cached Full Text

Cached at: 09/24/26, 03:38 AM

Paper page - Verifiable Hidden Dynamics Play: Generating Agentic RL Environments from Solved Mechanisms

Source: https://huggingface.co/papers/2609.27321

Abstract

Language-modelagentsincreasinglyfacelong-horizontaskswithevolvingstate,interdependentdecisions,anddelayedoutcomes.Scalingtheirtrainingrequiresdiverseagenticenvironments,dependableoutcomesignals,andlowextensioncost.Existinggenerationpipelinescommonlyconstructanenvironmentbeforedefiningitsoutcomeruleorannotatingitstrajectories,leavingdynamicsandevaluationtobealignedposthoc.VHD-Playreversesthisdependencybysamplingandsolvingamathematicalmodelbeforeacorpus-groundedsetterrendersitsdecisionprocessasstatefultools.Theexecutabledynamicsandtrajectory-scoringreferenceareinheritedfromthesamesolvedmodel.Thepipelineproduces3,300diverseagenticenvironmentsatacostofafewcentseach.TrainingQwen3.6-35B-A3Bonthreefamiliesraisesitsmeanagenticscorefrom0.204to0.815inafive-familydiagnostic.Gainsalsoappearonheld-outinstancesfromallthreetrainingfamiliesandeightunseenmechanismfamilies,thenextendbeyondthegeneratedsubstratetoexternalbenchmarksforgeneralfunctioncalling,travelplanning,and365-daye-commerce.OnE-CommerceBench,thetrainedcheckpointcompleteseveryrunwithoutbankruptcyandexceedsQwen3.7-Max.Wecomparewritten-outproblemswithstatefulversionsthatrevealorhidetheirparameters.Thecomparisonshowsthatmostofthelearnablegapliesinstatefulinteractionratherthanunderlyingproblemsolving.Afrozen35Bsetterrealizeslargerenvironments,andscale-matchedtrainingretainsgainsasmechanismsizeandhorizongrow,indicatingthepotentialforanevolvingtrainingsubstrate.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2609\.27321

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/2609.27321 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.27321 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.27321 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

AgentV-RL: Scaling Reward Modeling with Agentic Verifier

arXiv cs.CL

AgentV-RL introduces an Agentic Verifier framework that enhances reward modeling through bidirectional verification with forward and backward agents augmented with tools, achieving 25.2% improvement over state-of-the-art ORMs. The approach addresses error propagation and grounding issues in verifiers for complex reasoning tasks through multi-turn deliberative processes combined with reinforcement learning.