Implementation and Assessment of State Estimation Algorithms in Simulation and Real-World Applications

Abstract

State estimation algorithms are important mathematical tools for engineers and are capable of improving system modeling capabilities and scenario outcomes. Typically, systems utilize a variety of sensors to measure certain system states such as velocity or position; however, these sensors suffer from noise and biases which contaminate the states. Through implementing a state estimation algorithm, noise from low-cost sensors may be mitigated to provide better system state estimates. Thus, a need exists to apply these algorithms while using low-cost sensors and to assess the performance in different scenarios. The algorithms that were selected for analysis in this research were the Kalman Filter and Extended Kalman Filter, which were implemented in two separate experiments.

The first experiment encompassed a pursuer-evader scenario where the initial starting position of a pursuer and tracking measurement uncertainty of an evader were varied. Position of the evader was determined through two methods: raw tracking sensor measurements and estimates from a Kalman Filter. In both cases, the tracking sensor uncertainty was parameterized as a single term to represent combined uncertainty from all possible noise sources. This experiment showed that an increase in sensor measurement uncertainty led to an increase in the mean miss distance for the pursuer for both the raw tracking sensor method and Kalman Filter method. However, most engagement resulted in the Kalman Filter method providing an improved position estimate of the evader, reducing the average miss distance by upwards of 50%.

In the second experiment, an Extended Kalman Filter was applied to an aircraft that experienced a multitude of free-flight hardware failures such as control surface and aerial delivery failures. The Extended Kalman Filter was designed to estimate the aircraft’s stability and control derivatives along with the aircraft’s dynamic states. Fixed-position aileron failures, ranging from -30° to 30°, were assessed and showed a loss of effectiveness in the control derivative state estimates. An aerial delivery failure from a bay located near the center of gravity resulted in small changes in some of the control derivatives and noise characteristics of the aircraft; however, a payload release failure from a bay located on the wing resulted in every state estimate changing. Post-failure state estimate changes indicate the potential of implementing fault isolation control schemes to mitigate the failure after initial detection.

This research explored the usage of applying two state estimation algorithms to expand the modeling capabilities of two systems. Not only did the algorithms provide improved estimates for system states, the states that did not have direct sensor measurements were accurately estimated. Furthermore, the algorithms were tailored to each scenario and successfully utilized low-cost sensors to improve the scenario results.