environment-generation

AutoGym, a framework from Amazon AGI, generates complete reinforcement learning gyms—including tasks, executable environments, and verifiers—from minimal seeds using blueprint-first generation and active curriculum synthesis to support scalable agent training.

SkillGym proposes an automatic pipeline that crawls reproducible skills from the internet, builds verifiable difficulty-controlled environments, and collects 19k verified trajectories to train skill-use agents. Fine-tuning Qwen3.5 models (2B to 122B) on these trajectories improves performance across four skill-use benchmarks, with the 9B SFT model outperforming a 397B untrained model on two of them.

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.

The tweet highlights Odyssey-3's ability to generate visual environments for AI agents to learn from actions and consequences, enabling recursive intelligence, and references Google's Sima and Genie3 research.

EnvHarness introduces a programmable layer to dynamically reshape static environments for reinforcement learning, improving agent performance through automated targeting of weaknesses with EnvRigger.

SimWorld Studio is an open-source platform that uses an evolving coding agent to automatically generate and refine 3D environments for embodied agent learning. It leverages self-evolution and co-evolution mechanisms to create adaptive training scenarios, significantly improving agent performance.

ClawEnvKit is an automated pipeline that generates diverse, verified environments for claw-like agents from natural language descriptions, enabling the construction of Auto-ClawEval, a large-scale benchmark with 1,040 environments at 13,800x lower cost than human curation. The system supports continuous, on-demand evaluation and adaptive training environment generation across multiple model families and agent frameworks.