Researchers have developed a novel framework that integrates subjective evaluations from a Vision-Language Model (VLM) into a genetic algorithm for evolving virtual soft robots. By presenting the VLM with image sequences of robot locomotion and using subjective terms like 'adorably' and 'weirdly' for pairwise comparisons, the system guides the evolution of both morphology and locomotion. This VLM-driven selection accelerates population convergence compared to random selection and produces distinct phenotypes corresponding to the evaluation terms. An experiment with human participants indicated qualitative similarities in evolutionary outcomes, though with some individual choice divergence and noticeable fatigue from repeated evaluations. AI
IMPACT This research could lead to new methods in evolutionary computation and artificial life by enabling subjective linguistic expressions to shape embodied phenotypes.
RANK_REASON This is a research paper detailing a novel framework for evolutionary computation using VLMs for subjective evaluation.
Read on arXiv cs.NE (Neural & Evolutionary) →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
