ICANEWS

Engineering & Technology

Stochastic Differential Dynamic Programming for Partially Observable Trajectory Optimization

This study presents a stochastic differential dynamic programming algorithm designed for trajectory optimization challenges under partial observability. The met

Overview

This study presents a stochastic differential dynamic programming algorithm designed for trajectory optimization challenges under partial observability. The method addresses problems where trajectory design, orbit determination, and correction maneuver planning are interdependent. Its formulation considers a belief-state transition model and general mission constraints, with explicit accounting for how covariance propagation is influenced by the nominal trajectory, without relying on the separation principle.

Research Context

Designing spacecraft trajectories encounters difficulties due to stochastic effects, specifically 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 in uncertain environments. However, the application of practical methods for partially observable trajectory optimization, where multiple elements are tightly coupled, remains limited.

Approach

The proposed algorithm employs stochastic differential dynamic programming. It operates by optimizing both the nominal control sequence and feedback gains. A key aspect of the approach is its integration of a belief-state transition model and general mission constraints. The methodology differentiates itself by explicitly considering the relationship between covariance propagation and the nominal trajectory, rather than assuming independence as suggested by the separation principle.

Findings

Numerical examples were utilized to evaluate the algorithm's performance. These examples demonstrated that the proposed algorithm generated solutions that were both navigation-aware and robust to uncertainty. The evaluation encompassed a range of dynamical systems, observation models, and varying levels of uncertainty, indicating consistent performance across these conditions.

Why This Matters

The development of methods that can robustly design trajectories under partial observability and stochastic effects is relevant for applications requiring precise control and navigation in uncertain environments, such as spacecraft operations.

Key Findings