Ultrasonic Sensor Network: Target Localization with Passive Self-Localization

A flying robot is localized using ultrasonic sensor network distributed on the ground. This requires localization of sensor nodes first. In absence of RF link between target and sensors, weighted Multi-Dimensional Scaling (MDS) with Gradient Descent is used to localize the sensors from outlier free data obtained from generating a locally linear manifold. After initialization, an Extended Kalman filter (EKF) is used to estimate robot’s position alongwith other unknown parameters. Outliers, which are numerous because of large distances and noise involved, are detected by checking innovation to be within range. Variable length observation vector helps select the good sensor observations from the available many.