Advanced Control Methods for Autonomous Asteroid Exploration
Apollo 11 Postdoctoral Fellowships at Purdue - Proposal
Andrea Capannolo, Assistant Professor
Research Goals
Sustained and broad deep-space exploration is currently limited by the heavy reliance of spacecraft on ground-based operations. The human-in-the-loop decision-making process and the vast distances of celestial bodies impose fundamental constraints on spacecraft responsiveness and control capability. Developing autonomous systems can alleviate these constraints by reducing or eliminating dependence on ground commands. However, achieving full autonomy remains an open challenge. Many scenarios in space involve highly nonlinear dynamics and incomplete models, resulting in stringent requirements for both control precision and uncertainty management. Asteroids exemplify these challenges: their shapes and mass distributions are poorly characterized, producing irregular and uncertain gravitational fields that make proximity operations inherently complex and risky. Such nonlinear dynamics require accurate models to design and control trajectories effectively, yet reconstructing these unknowns to reduce uncertainty demands extensive data processing, an operation that exceeds the capacity of most onboard computers without heavy simplifications.
The overarching goal of this project is to develop advanced control strategies that enable autonomous operations in highly uncertain and nonlinear dynamical environments. The research integrates stochastic optimal control with high-order mathematical tools and machine learning techniques to move beyond linearized models and covariance-based uncertainty representations, offering a more accurate yet computationally tractable control framework under uncertainty.
Specifically, the proposed research will:
- Advance the use high-order mathematical methods such as Differential Algebra (DA) and State Transition Tensor (STT) to propagate nonlinear uncertainties in real time, allowing stochastic optimal control formulations to retain high-order dynamical information without compromising computational feasibility.
- Develop adaptive control algorithms that incorporate onboard learning of uncertain parameters through recurrent updates of the dynamical model.
- Establish a unified simulation platform to test proximity operations scenarios, including autonomous orbit insertion, hazard avoidance, and adaptive station-keeping under uncertain gravitational conditions.
By coupling learning-enabled modeling with high-order methods for optimal control, the project seeks to advance a new paradigm in deep-space autonomy, in which spacecraft progressively refine their understanding of the environment as they efficiently operate within it.
Expected Deliverables
Expected outcomes of the two-year appointment include:
- Publication of multiple journal and conference articles on the covered topics
- Dissemination of new developments at leading conferences
- Mentorship of graduate students working on related topics
- Preparation of grant proposals for funding extensions to the project
Affiliated Faculty
Assistant Professor Andrea Capannolo
School of Aeronautics and Astronautics, Purdue University