EvoPolicyGym: Evaluating Iterative RL Policy Self-Improvement by Coding Agents
University of Macau / CUHK
Introduces Autonomous Policy Evolution as an evaluation paradigm: a coding agent iteratively edits executable RL policy code, submits rollouts to a benchmark server, reads feedback, and refines — all under a fixed episode budget. Instantiated across 16 compact RL environments (Core-16). GPT-5.5 achieves the strongest aggregate rank score across all 16 environments.
Why it matters
Measures iterative self-improvement under budget constraints — a closer proxy for production agent deployments than one-shot task-success benchmarks.
Importance: 2/5
Addresses a gap in agent evaluation methodology with a reproducible testbed.
Sources
official
arXiv:2607.02440 — EvoPolicyGym
media
HuggingFace Daily Papers