Robust en-route and terminal navigation using topology and intensity returns from a forward-looking millimeter-wave radar

Abstract

In this paper we present Sierra Nevada Corporation's (SNC) Generalized Information Fusion Filter (GIFF). GIFF is a robust, sensor-agnostic estimation framework designed to blend measurements from a variety of sensors to produce an optimal estimate of the navigation state. At the core of GIFF is a Rao-Blackwellized (or marginalized) Particle Filter (RB-PF) with specialized Auxiliary Sampling Importance Resampling (ASIR). This algorithm places no limitation on the number of sensors it can use or on the linearity and error characteristics of their measurements, as opposed to more rigid, traditional techniques like Kalman Filters. This enables GIFF to process data from sensors of various kinds directly (3D radar/LIDAR, 2D surveillance radar, EO/IR, radar-altimeter, GPS, IMU, etc.), with minimal pre-processing. In addition, the marginalized implementation enables a large number of states to be estimated in real-time. We illustrate GIFF flexibility and performance using actual sensor data collected on fixed- and rotary-wing platforms equipped with an imaging radar producing 3D points and 2D images, a radar-altimeter, and an IMU. En-route tests show near-optimal accuracy is achieved during a one-hour flight over Virginia with a simulated GPS outage. GIFF is also initialized with large position uncertainty (5km) and shown to converge after only 30 seconds of flight. GIFF performance during terminal operations (landing) is illustrated using data collected on approaches to the Reno Stead airport, showing an accuracy similar to GPS 60 seconds before touchdown.