When Search Becomes Memory: Turning Robot Design Trials into Transferable Skills

2026-05-25

AI Summary

This work addresses the limitation of conventional large language model (LLM)-driven evolutionary robotics, which lacks a memory mechanism for historical experience and struggles to convert simulation outcomes into reusable knowledge. The authors propose Auto-Robotist, a self-evolving LLM agent that distills morphological search trajectories into a natural language skill library and retrieves relevant skills during evolution to guide the editing of elite individuals. The skill library is continuously updated through Add, Diagnose, and Merge operations. This approach represents the first explicit transformation of robot design search into auditable, transferable, structured linguistic knowledge, enabling a shift from implicit population memory to explicit design principles. Evaluated on seven EvoGym tasks, Auto-Robotist achieves substantially improved search efficiency in 5×5 design spaces under cold-start conditions and successfully transfers learned skills to 10×10 design spaces, outperforming pure genetic algorithms across all tasks.

Abstract

Large language models (LLMs) are increasingly used as proposal generators for evolutionary robot design, yet most loops remain memoryless: simulator results shape the next population but are not preserved as reusable design knowledge. We present Auto-Robotist, a self-evolving LLM agent that distills morphology-search traces into an explicit natural-language skill library. Each skill stores a structural archetype, evidence-grounded positive and negative rules, and the evaluated designs that support them, making design memory inspectable rather than implicit in a population. During search, the agent retrieves skills to condition LLM edits of elite bodies while retaining a Genetic Algorithm (GA) mutation path for exploration; after evaluation, it updates the library through Add, Diagnose, and Merge. Across seven EvoGym tasks spanning locomotion, traversal, and object interaction, Auto-Robotist improves cold-start 5x5 search and transfers learned skills to 10x10 design spaces, where reference-conditioned transfer outperforms GA on every task. These results suggest that LLM agents can convert expensive physical evaluations into reusable, auditable design principles. Our code will be released upon acceptance.

Problem

• robot design
• memoryless search
• transferable skills
• design knowledge
• large language models

Innovation

• LLM agent
• skill library
• transferable design knowledge
• evolutionary robotics
• memory-augmented search