依據觀測性分析建立自主式水下載具操控規劃器之研究

Abstract

觀測性為系統的初始狀態以及系統輸出之間的關係，本研究利用間接式同時定位與建地圖方法估測無人水下載具的狀態，此方法利用卡爾曼濾波器估測載具的慣性導航系統誤差並回授修正。利用此方法描述會使卡爾曼濾波器中的運動模型及觀測模型呈現線性，有助於分析系統的觀測性。首先，觀測性分析得出狀態空間內資訊含量為零的方向，並分析討論包含繞圈旋轉以及蛇行前進等運動所造成的狀態改變情況。接著利用這些操控動作結合協方差閥值建立載具操控規劃的方法，此方法在協方差大於閥值時會啟動觀測性操控動作以防止誤差繼續擴大。最後將這套方法實現於配有前視聲納之水下無人載具之方向控制，並模擬載具在二維空間的探勘情形。載具沿著設定好的路線移動，過程中斜方差大於閥值時即啟動觀測性操控動作，結果顯示在執行觀測性操控動作期間，載具加速度以及與特徵物間的位置向量會產生與直線前進時不同的變化，此變化造成資訊含量為零的方向改變，達成藉由操控載具使觀測性發生改變，並影響載具狀態準確度的目的。

Observability is an important property in linear system which relates the initial state and the system output. In this study, indirect SLAM using Kalman filter to estimate the error state of INS is used to describe the autonomous underwater vehicle state of motion. This method uses a linear time-varying equation to describe the AUV motion that enables the utilization of observability analysis during the AUV maneuver. Through the observability analysis, the unobservable mode which indicates the state direction without information from sensor observation in state space is identified. Then an observability-based maneuver planner is developed to prevent the state error from growing up. This maneuver planer involves a covariance threshold of heading and an observability-based maneuver patterns that involve heading maneuvers such as steady turn and S-shape motions. In this work, if the covariance is greater than the threshold, the maneuver planner will execute the observability-base maneuver to maintain the accuracy of states. An underwater vehicle which implemented the maneuver planner is simulated to explore a 2-D environment by a forward looking sonar . The vehicle explores in a set path and the covariance is tracked all the time. Once the covariance is greater than the threshold, the observability-based maneuver is executed. Simulation results show that variations of the accelerations of vehicle and the relative position between feature and vehicle are different between straight-line motion and heading turning or S-shape motions. The difference of variations of the projected length of unobservable mode on state space can also be shown. This indicates the unobservable mode can be controlled through maneuver and will affect the accuracy of vehicle state.