ICANEWS

Natural Sciences

Stochastic Differential Dynamic Programming for Partially Observable Trajectory Optimization

This paper presents a stochastic differential dynamic programming algorithm specifically designed for partially observable trajectory optimization problems. The

Overview

This paper presents a stochastic differential dynamic programming algorithm specifically designed for partially observable trajectory optimization problems. The methodology addresses challenges in spacecraft trajectory design influenced by stochastic effects, such as maneuver execution errors and observation uncertainties.

Research Context

Designing trajectories for spacecraft is complex when confronting stochastic elements like maneuver execution errors and observation uncertainties. Existing approaches, such as covariance control and belief-space planning, provide tools for developing robust control policies and information-aware trajectories under conditions of uncertainty. However, the prevalence of practical methods remains limited for trajectory optimization problems where trajectory design, orbit determination, and the planning of correction maneuvers are interconnected. These challenges point to a need for integrated solutions in partially observable scenarios.

Approach

The proposed method employs a stochastic differential dynamic programming algorithm. This algorithm is formulated to specifically address problems where trajectory design, orbit determination, and correction maneuver planning are tightly coupled. The core of the approach involves optimizing the nominal control sequence and the associated feedback gains. This optimization is conducted subject to a belief-state transition model. Additionally, the algorithm incorporates general mission constraints. A key aspect of this methodology is its explicit accounting for the dependence of covariance propagation on the nominal trajectory, without relying on the separation principle.

Findings

Numerical examples were used to evaluate the performance of the proposed stochastic differential dynamic programming algorithm. These examples demonstrated that the algorithm generated solutions characterized by both navigation awareness and robustness to uncertainty. This capability was observed across a range of different dynamical systems, various observation models, and differing levels of uncertainty within these simulated environments.

Why This Matters

The method provides a tool for designing trajectories which are robust against typical spacecraft operational uncertainties. It addresses coupled problems in trajectory optimization, where design, orbit determination, and maneuver planning interact directly. This could support the development of more reliable and precise trajectory designs for space missions.

Key Findings