Title:	Sim2Real REvolve using LLMs
DNr:	Berzelius-2025-17
Project Type:	LiU Berzelius
Principal Investigator:	Amy Loutfi <amy.loutfi@oru.se>
Affiliation:	Örebro universitet
Duration:	2025-01-27 – 2025-08-01
Classification:	10207
Homepage:	https://rishihazra.github.io/REvolve/
Keywords:

Abstract

Building on the success of REvolve, which demonstrated significant advancements in reward evolution for autonomous systems, this project extends its principles to Sim2Real learning for the Unitree Go2 robot. Utilizing NVIDIA’s Isaac platform, we aim to train policies across multiple simulation environments with extensive GPU resources, enabling robust and scalable solutions. By leveraging TensorFlow GPU and lessons from REvolve, we will ensure efficient training, faster convergence, and seamless transfer to real-world testing. This continuation of REvolve focuses on enhancing adaptability, scalability, and performance in dynamic environments, bridging the gap between simulation and reality.