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

Excerpt

arXiv:2609.27321v1 Announce Type: cross Abstract: 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