Small satellites like CubeSats are attractive for deep-space science because they are cheap and compact, but their size forces hard tradeoffs. Attitude control—knowing and adjusting which way the spacecraft points—is a major one. Many small satellites rely on magnetorquers, which use Earth's magnetic field to generate torque, but these provide only weak and sometimes underactuated control, especially far from Earth.
A new preprint on arXiv proposes a learning-based adaptive model predictive control (MPC) scheme for magnetorquer-only attitude control. The idea is to combine the constraint-handling strengths of MPC with an adaptive component that learns the spacecraft's behavior over time, potentially improving pointing performance without adding heavy hardware like reaction wheels.
The work is aimed at nanosatellites intended for outer solar system exploration, where every kilogram counts and where traditional attitude control actuators may be impractical. If the approach holds up, it could give small probes a more capable and reliable way to orient themselves for science observations, though the paper's abstract does not yet provide detailed simulation or flight results.