Automated Design Competition Technical Report: Cascaded Structure and Parameter Optimization Based on Prior Knowledge

Xiang Shu, Yiyi Zhu, Renji Zhang, Xiang Xia, Bingdong Li, Hong Qian

Abstract: The Automated Design Competition in GECCO'24 aims to find intelligent 3D agents that perform better in specific environments. To address this problem, this technical report proposes a Cascaded Structure and Parameter Optimization (CaSPO) framework. After constructing an initial population by using prior knowledge, CaSPO optimizes the mechanical structure, control system, and parameters sequentially to find good-performance agents. Experiments in different tasks and environments verify the effectiveness of the proposed CaSPO framework.