CIRC COSTA: An Open Cloud Simulation Platform for Physical Intelligence

The project aims to develop COSTA, an innovative open-source computational infrastructure designed to revolutionize how researchers in Computer and Information Science and Engineering (CISE) approach physical simulation and physics-based learning. By creating a modern, cloud-native, and fully differentiable platform, the project seeks to address the current challenges researchers face when implementing complex numerical methods and physical simulations, which often require significant expertise and computational resources.

COSTA will provide unprecedented accessibility to cutting-edge computational techniques, enabling researchers across various domains—including robotics, computer vision, deep learning, and computational physics—to leverage advanced simulation and learning tools. The platform's key innovations include optimal utilization of multi-node computing architectures, full differentiability for seamless integration into learning pipelines, and a cloud-hosted environment that allows easy and cost-effective simulation runs. Importantly, the project emphasizes community governance, with a long-term goal of creating a sustainable organization to maintain and develop the platform, ultimately democratizing access to sophisticated computational methods for researchers from diverse institutional backgrounds.