Gym-V: A Unified Vision Environment System for Agentic Vision Research

TL;DR

Gym-V is a comprehensive platform of 179 visual environments designed to facilitate systematic research on agentic vision systems, highlighting the importance of observation scaffolding and diverse training for effective learning.

Contribution

Introduces Gym-V, a unified, large-scale visual environment system enabling controlled experiments and cross-domain transfer studies for agentic vision research.

Findings

Observation scaffolding is more critical than RL algorithm choice.

Captions and game rules determine learning success.

Training on diverse tasks generalizes broadly, narrow training can cause negative transfer.

Abstract

As agentic systems increasingly rely on reinforcement learning from verifiable rewards, standardized ``gym'' infrastructure has become essential for rapid iteration, reproducibility, and fair comparison. Vision agents lack such infrastructure, limiting systematic study of what drives their learning and where current models fall short. We introduce \textbf{Gym-V}, a unified platform of 179 procedurally generated visual environments across 10 domains with controllable difficulty, enabling controlled experiments that were previously infeasible across fragmented toolkits. Using it, we find that observation scaffolding is more decisive for training success than the choice of RL algorithm, with captions and game rules determining whether learning succeeds at all. Cross-domain transfer experiments further show that training on diverse task categories generalizes broadly while narrow training can cause negative transfer, with multi-turn interaction amplifying all of these effects. Gym-V is released as a convenient foundation for training environments and evaluation toolkits, aiming to accelerate future research on agentic VLMs.