TL;DR
This paper introduces a State-Dependent Riccati Equation (SDRE) control approach for guiding spacecraft during the hypersonic phase of atmospheric entry, specifically for a Mars mission. By leveraging sum-of-squares programming, the SDRE method offers a theoretically sound and practically applicable way to optimize control effort and achieve precise landing targets, outperforming a traditional guidance algorithm in certain scenarios.
What Problem Does This Paper Address?
Accurately landing spacecraft on celestial bodies like Mars is becoming increasingly challenging as desirable landing sites become scarcer and mission requirements demand higher precision. Traditional guidance systems for atmospheric entry, like the Modified Apollo Final Phase (MAFP) algorithm, often rely on pre-defined reference trajectories and don't explicitly optimize for control effort (e.g., propellant usage). This can lead to suboptimal performance, especially when precise state targeting (position and velocity) is critical. The paper addresses the need for a guidance algorithm that can handle the highly nonlinear dynamics of atmospheric entry, optimize performance, and be suitable for onboard implementation, while also considering the trade-offs between accuracy and propellant consumption.
Methods
The core of the paper is the application of the State-Dependent Riccati Equation (SDRE) control technique. SDRE is an emerging method for controlling nonlinear systems by factorizing their dynamics into a linear-like structure with state-dependent matrices. This allows for the application of Linear Quadratic Regulator (LQR) principles, adapted for nonlinear systems. The authors extend this by using sum-of-squares (SOS) programming to solve the SDRE. SOS programming is a powerful tool for verifying polynomial inequalities and finding polynomial solutions, which is particularly useful for synthesizing control laws that guarantee stability and robustness.
Key methodological steps include:
- System Factorization: Representing the nonlinear entry dynamics in a state-dependent coefficient (SDC) form: $\dot{x}(t) = A(x)x(t) + B(x)u(t)$.
- SDRE Formulation: Deriving a state-dependent algebraic Riccati equation (SDRE) from the Hamilton-Jacobi-Bellman (HJB) theory, analogous to the LQR problem.
- Sum-of-Squares Synthesis: Employing SOS programming to find polynomial solutions for the SDRE, which can then be used to derive the state-feedback control law $u(x) = -K(x)x$. This approach allows for the synthesis of controllers that satisfy stability conditions.
- Reference Trajectory Tracking: Adapting the SDRE to track a pre-defined reference trajectory by defining an error system.
Datasets
The paper does not introduce new datasets. Instead, it utilizes a simulated high-mass Mars entry vehicle model based on established physics and parameters specific to Mars. This model includes kinematic and dynamic equations governing the vehicle's descent, atmospheric density, and gravitational field. Performance is evaluated against a reference trajectory defined by initial and final conditions, and compared to the Modified Apollo Final Phase (MAFP) algorithm, which is a flight-heritage guidance law.
Key Insights and Ideas
- SDRE as a Nonlinear LQR Extension: The SDRE provides a principled way to extend LQR concepts to nonlinear systems, offering a more direct path to optimal control compared to traditional methods that might linearize around a trajectory.
- Sum-of-Squares for Robust Control Synthesis: Using SOS programming to solve the SDRE is a novel application that allows for the synthesis of provably stable and robust controllers, especially for polynomial systems. This moves beyond numerical solutions to analytical guarantees.
- Propellant Optimization: A significant advantage of the SDRE approach is its ability to directly account for control effort (propellant usage) in the cost function, leading to more fuel-efficient trajectories compared to guidance laws that don't explicitly optimize this metric.
- Full-State vs. Partial-State Tracking: The SDRE approach tracks the full state of the system, offering potentially higher precision than MAFP, which tracks a subset of states (range-to-go, acceleration, altitude rate, velocity).
- Region of Attraction Analysis: The paper demonstrates the ability to estimate the region of attraction for the SDRE controller, providing confidence in its performance within a defined operational envelope.
Key Results
The paper evaluates the SDRE-based guidance algorithm against the MAFP algorithm in two scenarios:
- 10% Higher Density:
- Velocity Error at 7 km: SDRE (Continuous) 7.5 m/s, Sum-of-Squares SDRE 9.4 m/s, MAFP 8.6 m/s.
- Position Error at 7 km: SDRE (Continuous) 420 m, Sum-of-Squares SDRE 440 m, MAFP 398 m.
- Propellant Usage: SDRE (Continuous) 27.6 kg, Sum-of-Squares SDRE 29.2 kg, MAFP 33.9 kg.
- Shallow Flight Path Angle ($\gamma = 8^\circ$):
- Velocity Error at 7 km: SDRE (Continuous) 9.6 m/s, Sum-of-Squares SDRE 29.6 m/s, MAFP 16.3 m/s.
- Position Error at 7 km: SDRE (Continuous) 378 m, Sum-of-Squares SDRE 622 m, MAFP 287 m.
- Propellant Usage: SDRE (Continuous) 27.6 kg, Sum-of-Squares SDRE 31.2 kg, MAFP 28.9 kg.
Qualitative Findings: * The SDRE approach, particularly the continuously updated version, generally shows comparable or better accuracy than MAFP while achieving significant propellant savings. * The sum-of-squares implementation of SDRE performed slightly poorer than the continuous SDRE, which is attributed to the update rate of the local solution. * All algorithms showed similar behavior in reducing relative error, with early error growth primarily dominated by timeline differences.
Limitations and Future Work
- Taylor Series Approximation: The extended linearization of the system relies on Taylor series approximations for trigonometric and exponential functions. The paper acknowledges that the impact of these approximations on the solution quality needs further investigation.
- Local vs. Global Optimality: The SDRE approach, especially when solved using SOS programming with local updates, provides a locally optimal solution. While it offers asymptotic stability, global asymptotic stability is harder to guarantee and depends on specific conditions.
- Computational Cost: While SOS programming offers analytical guarantees, its computational cost can be significant, especially for high-dimensional systems or complex polynomial representations. The paper mentions using external solvers like SOSTOOLS and SeDuMi.
- Future Work: The authors suggest investigating alternative extended linearizations of the system and further characterizing the impact of Taylor series approximations. They also propose exploring the trade-offs between different SDRE solution techniques and their computational demands.
Summary updated