Abstract
This article extends the optimal covariance steering (CS) problem for linear stochastic systems subject to chance constraints so as to account for an optimal allocation of the risk. Previous works have assumed a uniform risk allocation in order to cast the optimal control problem as a semidefinite program, which can be solved efficiently using standard SDP solvers. An iterative risk allocation (IRA) formalism is used to solve the optimal risk allocation problem for CS using a two-stage approach. The upper-stage of IRA optimizes the risk, which is a convex problem, while the lower-stage optimizes the controller with the new constraints. The process is applied iteratively until the optimal risk allocation that achieves the lowest total cost is obtained. The proposed framework results in solutions that tend to maximize the terminal covariance, while still satisfying the chance constraints, thus leading to less conservative solutions than previous methodologies. In this article, we consider both polyhedral and cone state chance constraints. Finally, we demonstrate the approach to a spacecraft rendezvous problem and compare the results with other competing approaches.
Introduction
In stochastic control, managing systems influenced by random noise—such as atmospheric turbulence for drones or sensor noise for spacecraft—is a fundamental challenge. Since Gaussian noise is theoretically unbounded, absolute safety cannot be guaranteed. Instead, researchers use chance constraints, which ensure the probability of failure remains below a specified threshold (e.g., a 97% safety guarantee).
Traditionally, the total risk budget ($\Delta$) is distributed uniformly across all time steps and constraints. However, this "one-size-fits-all" approach is often inefficient. Some maneuvers are inherently safer than others; forcing a uniform risk level everywhere makes the controller overly cautious, leading to higher fuel consumption and restricted operational envelopes. This paper introduces a method to treat risk as a resource to be optimized, allowing for more efficient and less conservative autonomous systems.
The Iterative Risk Allocation (IRA) Framework
The core contribution of this work is a two-stage Iterative Risk Allocation (IRA) algorithm. Rather than fixing risk levels, the IRA treats them as decision variables to be optimized alongside the control policy.
As detailed in Algorithm 1, the process involves:
- Upper Stage (Risk Optimizer): This stage treats the risk levels ($\delta$) for each constraint as decision variables. It uses a convex optimization approach to adjust these levels based on the current controller's performance.
- Lower Stage (Controller Optimizer): Given the risk levels from the upper stage, this stage solves a Semi-Definite Program (SDP) to find the optimal control gains that steer the state mean and covariance to the target.
This iterative process continues until it identifies the optimal risk distribution that minimizes the total control cost while strictly adhering to the global risk budget.
Handling Complex Geometric Constraints
Many real-world constraints, particularly in aerospace, are conical (e.g., a camera's field of view or a docking port's approach zone). These are mathematically difficult to handle while maintaining the convexity required for efficient solving.
The authors propose a novel Geometric Approximation for 3D cone constraints. By leveraging the properties of Gaussian distributions to create a "disk" of safety, this method proved to be significantly less conservative than traditional linear relaxations or probabilistic union bounds. This allows the spacecraft to operate closer to the boundaries of its safe zone when beneficial for fuel efficiency.
Key Results and Insights
The proposed framework was evaluated using a high-fidelity simulation of a Spacecraft Rendezvous and Docking mission governed by Clohessy-Wiltshire-Hill equations. Key findings include:
- Efficiency and Cost Reduction: The IRA method utilized nearly the entire 3% risk budget (achieving a true risk of ~2.99%), whereas uniform allocation was "too safe," reaching only ~1.23%. By not fighting to stay unnecessarily far from boundaries, the IRA method achieved a lower optimal control cost.
- Maximizing Terminal Covariance: A less conservative solution allows for a larger terminal covariance. This indicates that the controller does not need to expend energy to over-constrain the spacecraft's position, resulting in significant fuel savings.
- Superiority over SMPC: Unlike standard Stochastic Model Predictive Control (SMPC), which often struggles with recursive feasibility and precise steering of final uncertainty, the IRA method handles both effectively, making it ideal for precision tasks like orbital docking.
Conclusion
By shifting the perspective of risk from a static constraint to an optimizable resource, this research provides a path toward more capable autonomous systems. While currently focused on Linear Time-Varying (LTV) systems with Gaussian noise, the framework demonstrates that intelligent risk allocation is key to reducing conservatism and improving performance in high-stakes stochastic environments.
Summary updated