Uncertainty-Aware Path Planning for Stewart Platforms

Abstract

Stewart platforms are widely used in precision aerospace applications due to their high stiffness and six-degree-of-freedom actuation capabilities. In tasks such as in-orbit assembly and robotic manipulation, maintaining accurate knowledge of the platform state throughout motion is critical. Traditional sampling-based motion planners typically optimize geometric objectives such as path length, which does not directly account for the uncertainty that accumulates during actuation. This work presents an uncertainty-aware motion planning framework for Stewart platforms that incorporates belief propagation using an Extended Kalman Filter (EKF). The EKF is used to propagate state covariance along candidate trajectories, enabling the trace of the error covariance matrix to be used as a cost function during planning. Sampling-based motion planners from the Rapidly-Exploring Random Tree (RRT) family are evaluated within the full six-dimensional platform configuration space. Comparative experiments show that RRT# demonstrates superior convergence behavior relative to RRT and RRT*. When uncertainty is used as the optimization metric, the planner successfully identifies trajectories with progressively lower estimation uncertainty over time. Experimental results further indicate that path length and path uncertainty are not strongly correlated, suggesting that minimizing estimation uncertainty may require trajectories that differ from geometrically optimal paths. These results demonstrate the potential of uncertainty-aware sampling-based planning for improving reliability in high-precision robotic systems.