TLDR: This research introduces an Explainable AI-enhanced supervisory control framework for high-precision spacecraft formation flying, specifically for the Virtual Telescope for X-ray Observation (VTXO) mission. It combines timed automata for mission phase management, Sliding Mode Control (SMC) for robust relative position and attitude control, and Deep Neural Networks (DNN) for predicting optimal mission parameters and performance metrics (energy and error). The framework ensures transparency and trustworthiness in control decisions, validated through extensive simulations, to achieve sub-millimeter alignment and 55 milli-arcsecond angular resolution for X-ray observations.
Space exploration is pushing the boundaries of precision, especially with missions like the Virtual Telescope for X-ray Observation (VTXO). This ambitious project involves two separate spacecraft flying in a precise formation to create a virtual telescope with a one-kilometer focal length. The goal is to observe high-energy space objects with an unprecedented 55 milli-arcsecond angular resolution, which is significantly finer than existing X-ray telescopes.
Achieving such extreme precision in space is no small feat. It requires maintaining a sub-millimeter transverse alignment between the two spacecraft over a vast distance, all while managing dynamic uncertainties, disturbances, and strict energy budgets. Traditional control systems often struggle with these complex, real-time demands.
A New Approach with AI and Adaptive Control
Researchers have introduced a sophisticated framework that combines artificial intelligence (AI) with supervisory adaptive control systems to tackle these challenges. The core idea is to optimize mission planning and execution, enhancing the efficiency and accuracy of the VTXO mission. This framework integrates several advanced techniques:
- Timed Automata for Supervisory Control: This acts as the mission’s brain, managing the various operational phases of the spacecraft. It ensures smooth and logical transitions between stages like formation stabilization, transient maneuvers, and the critical science observation phase.
- Sliding Mode Control (SMC): For the high-precision tasks of maintaining relative position and attitude during science observations, SMC is employed. Known for its robustness, it helps the spacecraft stay on target even when faced with external disturbances and sensor noise.
- Explainable AI (XAI): A crucial element for safety-critical space missions, the XAI component provides transparency. Instead of just giving commands, it predicts key performance metrics like energy consumption and mission error for a given set of parameters. This allows mission operators to understand the trade-offs involved in control decisions, fostering trust and enabling real-time adjustments.
- Neural Network-Based Optimization: Deep neural networks are used to solve complex, constrained dynamic optimization problems. These networks learn from extensive offline simulations to predict optimal mission parameters, ensuring that precision criteria are consistently met.
Mission Phases and Control Strategies
The VTXO mission is broken down into distinct phases for both its Attitude Control System (ACS) and relative position formation. The ACS has three phases: formation, transient, and science. The position formation has four phases: de-formation, tracking, formation, and science. Each phase has specific objectives and control strategies. For instance, during the transient phase, a Lyapunov controller guides the spacecraft to the desired alignment, while the science phase relies on the robust SMC for precise formation flying.
Navigation is critical, utilizing a suite of sensors including Inertial Measurement Units (IMU), star trackers (like NISTEx-II), laser beacons, GPS (for lower altitudes), and radio ranging. The Multiplicative Extended Kalman Filter (MEKF) processes data from these sensors to provide accurate state estimations, crucial for the control systems.
The AI Framework in Action
The AI framework operates in three stages: an offline optimization and data production phase, an offline machine learning phase, and a real-time implementation phase. In the first stage, powerful optimization algorithms like Simulated Annealing (SA) and Multi-objective Genetic Algorithm (MOGA) are used to find optimal control parameters that minimize energy consumption and pointing error. This generates a rich dataset.
In the second stage, a Deep Neural Network (DNN) is trained on this dataset. The DNN learns to predict the optimal controller parameters and performance metrics (energy and error) based on the spacecraft’s initial and desired states. This training is done offline, leveraging the computational power available on Earth.
Finally, in the real-time implementation phase, the trained DNN is deployed on hardware like FPGAs aboard the spacecraft. It takes real-time sensor data and quickly outputs the optimal control parameters, enabling adaptive and efficient maneuvers. The explainable nature of this AI means that operators can always see the predicted impact of control decisions on energy and error, making the system transparent and trustworthy.
Also Read:
- Bridging AI and Space: ASTREA’s Agentic Approach to Orbital Thermal Autonomy
- Closing the Gap: Improved Spacecraft Pose Estimation for Real-World Missions
Ensuring Reliability and Performance
Extensive verification and validation (V&V) processes, including Monte Carlo simulations and formal stability arguments, confirm the system’s accuracy, reliability, and efficiency. These tests demonstrate that the VTXO GNC architecture can meet its stringent precision, energy, timing, and robustness requirements under realistic space conditions. The use of SMC, for example, has been shown to achieve sub-millimeter transverse alignment, outperforming simpler proportional-derivative (PD) controllers.
This research represents a significant step forward in autonomous spacecraft control, offering a blueprint for future high-precision space missions. By combining robust control techniques with an explainable AI, the VTXO mission is poised to unlock new discoveries in X-ray astronomy. You can find more details about this research at arXiv:2509.13331.
