A skill library trained on 5x5 robot grids beats genetic search at 10x10 — on all 7 EvoGym tasks

Notes on When Search Becomes Memory: Turning Robot Design Trials into Transferable Skills (arXiv:2605.25832) — Yunfei Wang, Xiaohao Xu, Yang Li, Xiaonan Huang · May 2026

Note published July 24, 2026 · written by SkillFed’s research pipeline from the paper above · how these notes are made

Auto-Robotist is a self-evolving LLM agent for evolutionary robot design that turns each generation's simulator results into a structured, natural-language skill library instead of discarding them once the next population is bred. Each entry pairs a structural archetype — a recurring body-plan pattern — with evidence-grounded positive and negative rules, anchored to the actual evaluated designs that produced them, so the memory stays inspectable rather than folding into model weights or a population's genome. During search, the agent retrieves relevant skills to steer LLM edits of elite robot bodies, while keeping a standard genetic algorithm mutation path open for exploration. After each evaluation round, the agent maintains that library itself — adding skills earned from the newest evidence, catching and correcting ones that no longer hold, merging entries that have drifted into duplicates.

Across seven EvoGym tasks spanning locomotion, terrain traversal, and object interaction, the skill-conditioned approach improves cold-start search on small 5x5 design grids. Skills learned at that scale go on to transfer: at a much larger 10x10 design space, reference-conditioned search beats a plain genetic algorithm on every one of the seven tasks. The result reframes what an expensive simulated evaluation is for — not a one-off score used to cull a population, but evidence banked into an auditable, reusable design memory that keeps paying off at a scale the agent never searched directly.