Modular Multi-Fidelity Simulation Framework for Advanced Sail-Driven Spacecraft Control and Dynamics in Perturbation-Rich Regimes

This work presents a high-fidelity Software-in-the-Loop (SIL) simulation framework for the development and validation of control strategies for sail-driven spacecraft operating in Low Earth Orbit (LEO). The framework integrates NVIDIA Omniverse with the Robot Operating System (ROS) to enable physics-based real-time simulation, photorealistic rendering, and seamless hardware integration, following a digital twin paradigm. Orbital dynamics modeling includes key LEO perturbations such as gravity gradients, gravitational anomalies, and relative motion effects to ensure realistic operational scenarios. A central contribution is a multi-fidelity atmospheric drag modeling architecture that applies forces directly to the spacecraft geometry through the physics engine, ensuring an accurate and design-agnostic representation of aerodynamic interactions. Two complementary computation modes are implemented: (i) a high-fidelity, sensor-based, model-agnostic approach relying on virtual raycasting for precise wrench estimation on the spacecraft surfaces; and (ii) a computationally efficient Feed-Forward Neural Network (FFNN) surrogate model that approximates the nonlinear aerodynamic response for the trained range of attitudes and orbital conditions. The sensor-based method provides high fidelity for detailed analysis and validation, while the FFNN surrogate enables high-throughput simulations for large-scale Reinforcement Learning (RL) campaigns involving thousands of parallel environments. This combination supports rapid prototyping and benchmarking of both classical and learning-based guidance, navigation, and control algorithms for solar- and drag-sail spacecraft, ultimately reducing development risk and accelerating the deployment of fuel-free, extended-lifetime LEO missions.