CUDA-Accelerated Soft Robot Neural Evolution with Large Language Model Supervision
Lechen Zhang
TL;DR
This paper tackles the co-design of soft robot morphology and control, a problem confounded by complex morphology–control interactions that traditional approaches treat as separate steps. It presents an integrated framework that encodes morphology and control in a single-genome MLP and evolves it with a CUDA-accelerated simulator, augmented by Gaussian positional encoding and LLM supervision. The key contributions are (i) an implicit dual-encoding scheme within the network, (ii) high-throughput GPU-accelerated evolution reaching up to about 7.9 billion simulations per second on a single RTX-3090, (iii) Gaussian positional encoding to enhance morphology understanding, and (iv) LLM-based guidance for faster convergence while preserving diversity. The results demonstrate improved design efficiency and broader exploration of the design space, enabling more capable and rapidly discovered soft-robot configurations.
Abstract
This paper addresses the challenge of co-designing morphology and control in soft robots via a novel neural network evolution approach. We propose an innovative method to implicitly dual-encode soft robots, thus facilitating the simultaneous design of morphology and control. Additionally, we introduce the large language model to serve as the control center during the evolutionary process. This advancement considerably optimizes the evolution speed compared to traditional soft-bodied robot co-design methods. Further complementing our work is the implementation of Gaussian positional encoding - an approach that augments the neural network's comprehension of robot morphology. Our paper offers a new perspective on soft robot design, illustrating substantial improvements in efficiency and comprehension during the design and evolutionary process.