TL;DR
This paper introduces a trajectory generation method for rudderless, dual vectored-thrust flying wings, a platform where pitch control and translation are inherently coupled. By jointly optimizing path geometry and timing through trajectory inversion, the authors enable complex aerobatic maneuvers while significantly reducing actuator demand and ensuring dynamic feasibility.
What Problem Does This Paper Address?
Flying wings are aerodynamically efficient and compact, but removing traditional control surfaces (like rudders and elevators) creates a significant control challenge. In the dual vectored-thrust configuration studied here, the aircraft relies entirely on two tilting propulsors for both propulsion and attitude control.
The core technical hurdle is force-moment coupling: the thrust component required to generate a pitching moment also contributes to the aircraft's translational acceleration. Consequently, you cannot simply prescribe a geometric path and an attitude independently; they are physically linked. Generating "dynamically compatible" trajectories—paths the aircraft can actually follow without saturating its actuators or violating physics—remains an open challenge for this class of UAV.
Methods
The authors propose a two-step approach to trajectory generation:
- Trajectory Inversion for Attitude Recovery: The researchers developed a method to recover the required attitude history along a prescribed path. By solving "compatibility equations" derived from the aircraft's force and moment balances, they ensure that the recovered attitude and propulsor commands satisfy the inherent coupling of the rudderless design.
- Joint Geometry and Timing Optimization: The path is represented by quintic B-splines, while a second B-spline controls the "progress rate" (timing). The study uses a coordinate search to optimize both simultaneously. The objective function balances peak acceleration against aerodynamic support, aiming to lower the demand on the thrust-vectoring system while preserving the maneuver's defining features.
For evaluation, the generated trajectories are tracked in a high-fidelity nonlinear simulation using a combined Nonlinear Model Predictive Control (NMPC) and Incremental Nonlinear Dynamic Inversion (INDI) controller.
Datasets
The study does not use external datasets. Instead, it utilizes a high-fidelity mathematical model of a 0.405 kg flying wing. The aerodynamic model is based on wind-tunnel-derived periodic coefficient fits, which allow the simulation to remain accurate even during the large-angle excursions typical of aerobatic flight.
Key Insights and Ideas
- Timing as a Control Variable: A key "aha" moment is that adjusting when the aircraft moves is as vital as where it moves. Slowing down in a tight curve reduces normal acceleration but also reduces aerodynamic lift, which might actually increase the demand for thrust vectoring. The joint optimization finds the optimal trade-off.
- Non-Algebraic Attitude: Unlike some simpler UAV models, a position-heading history alone does not algebraically determine the attitude for this aircraft. The internal attitude motion must be solved as a differential equation (the compatibility equation) to ensure the force-moment balance is maintained.
- Actuator Relief: The optimization process significantly smooths out actuator demands. In an S-shaped climb, joint optimization reduced the peak vectoring angle from 17.2° to 10.4° and the vectoring rate from 171°/s to just 46°/s.
Key Results
The method was validated across three representative aerobatic sequences:
- Cloverleaf (F1): Successfully performed four lobes with quarter-rolls, maintaining a peak load of only 1.34g and a tiny position tracking error of 1.3 mm.
- Half-Cuban (F4): Demonstrated climbing, inversion, and recovery. This tighter maneuver required a higher specific-force load (2.94g) but remained well within the aircraft's limits.
- Straight Rolling (F8): Showed that the method can separate geometric pathing from required body-axis rotations, successfully executing continuous and four-point rolls while maintaining a linear flight path.
Limitations and Future Work
- Simulation-Based: While the aerodynamic models are grounded in wind-tunnel data, the results are currently limited to numerical simulations. Real-world factors like gusty winds, sensor noise, and communication latency were not the primary focus.
- Computational Complexity: The joint optimization uses a coordinate search on B-spline control points. While effective, the authors do not discuss if this method is fast enough for real-time, on-board replanning during flight.
- Future Directions: The authors suggest this work provides a foundation for fully autonomous aerobatics. Future research will likely involve transitioning these algorithms to physical hardware for field testing.
Summary updated