Session
Flash Talks Session 3
Location
Salt Palace Convention Center, Salt Lake City, UT
Abstract
Spacecraft attitude control traditionally requires extensive per-mission engineering—hand-tuned gains, mission-specific mode logic, and conservative operating envelopes—that breaks down when on-orbit conditions deviate from ground assumptions. This is especially costly for small satellites, where limited ground support makes autonomous adaptability essential. We present a feasibility-aware attitude trajectory planner that replaces this hand-engineered stack with continuous planning from a digital twin: less a new feedback law than a different division of labor, in which the mode logic, gain tuning, and desaturation a conventional ADCS builds as separate pieces are instead absorbed into a single optimization.
Built on the ALTRO trajectory optimization algorithm, the planner accepts high-level pointing goals and reasons over future time-varying control authority, respecting the full nonlinear dynamics and actuator constraints directly—without collocation or slack variables. Because trajectories are produced by forward integration, every returned trajectory is dynamically consistent with the modeled spacecraft; when a goal cannot be met within the modeled constraints, the planner returns a bounded, best-effort trajectory rather than exceeding actuator authority or diverging. It also emits time-varying LQR gains for closed-loop tracking, providing robustness to unmodeled disturbances, estimation error, and sensor noise without anticipating them at the planning stage.
Optimizing directly against the dynamics, the planner also finds strategies a designer would otherwise have to anticipate: facing a body-fixed disturbance beyond its actuator authority, it autonomously adopted a spinning maneuver that averages the disturbance out over each revolution—a solution it was never directed toward. Monte Carlo simulations (100 runs, 1000s, randomized orientation, goal, rate, and orbit, under disturbances excluded from the planner’s model) show 84% of slews converging ( < 5 degrees, mean 3.8 degrees) for a magnetorquer-only spacecraft versus 27% (mean 21.5 degrees) for PD control, improving to 94% (mean 1.2 degrees) versus 90% (mean 9.1 degrees) with a single reaction wheel added.
The same framework spans magnetorquer-only, hybrid, and full reaction-wheel configurations with minimal reconfiguration, handling momentum management and desaturation implicitly within slews and pointing. Implemented in C++ for embedded-class processors, it replans continuously on overlapping horizons—bringing trajectory-planned autonomy to spacecraft historically limited to reactive control.
Document Type
Event
Included in
A Feasibility-Aware Attitude Trajectory Planner for Small Satellites
Salt Palace Convention Center, Salt Lake City, UT
Spacecraft attitude control traditionally requires extensive per-mission engineering—hand-tuned gains, mission-specific mode logic, and conservative operating envelopes—that breaks down when on-orbit conditions deviate from ground assumptions. This is especially costly for small satellites, where limited ground support makes autonomous adaptability essential. We present a feasibility-aware attitude trajectory planner that replaces this hand-engineered stack with continuous planning from a digital twin: less a new feedback law than a different division of labor, in which the mode logic, gain tuning, and desaturation a conventional ADCS builds as separate pieces are instead absorbed into a single optimization.
Built on the ALTRO trajectory optimization algorithm, the planner accepts high-level pointing goals and reasons over future time-varying control authority, respecting the full nonlinear dynamics and actuator constraints directly—without collocation or slack variables. Because trajectories are produced by forward integration, every returned trajectory is dynamically consistent with the modeled spacecraft; when a goal cannot be met within the modeled constraints, the planner returns a bounded, best-effort trajectory rather than exceeding actuator authority or diverging. It also emits time-varying LQR gains for closed-loop tracking, providing robustness to unmodeled disturbances, estimation error, and sensor noise without anticipating them at the planning stage.
Optimizing directly against the dynamics, the planner also finds strategies a designer would otherwise have to anticipate: facing a body-fixed disturbance beyond its actuator authority, it autonomously adopted a spinning maneuver that averages the disturbance out over each revolution—a solution it was never directed toward. Monte Carlo simulations (100 runs, 1000s, randomized orientation, goal, rate, and orbit, under disturbances excluded from the planner’s model) show 84% of slews converging ( < 5 degrees, mean 3.8 degrees) for a magnetorquer-only spacecraft versus 27% (mean 21.5 degrees) for PD control, improving to 94% (mean 1.2 degrees) versus 90% (mean 9.1 degrees) with a single reaction wheel added.
The same framework spans magnetorquer-only, hybrid, and full reaction-wheel configurations with minimal reconfiguration, handling momentum management and desaturation implicitly within slews and pointing. Implemented in C++ for embedded-class processors, it replans continuously on overlapping horizons—bringing trajectory-planned autonomy to spacecraft historically limited to reactive control.
