Arquitetura systolic array baseada em FPGA para o filtro de Kalman robusto aplicado a um sistema aéreo

Abstract

This dissertation presents a proposed implementation of a hardware accelerator, with parallel architecture based on systolic array, for building the navigation system in an aerial vehicle. The estimate of the attitude, position, and speed of the drone is established by a robust Kalman filter due to possible uncertainties in the parameters of the sensors. The structure is assisted by data from an inertial measurement unit, attitude, and position information from a monocular visual SLAM solution, and position and velocity references obtained via GPS. Furthermore, the operating modes of the navigation system are defined according to the availability and reliability of the sensors. Thus, the flexibility of the air vehicle operation in different environments will depend on the multiplicity and quality of the information sources to estimate the states of the air navigation system. The main contribution of this study is the possibility of building a state estimation system with accurate and real-time results, which is a fundamental need in aerial robotics. For this, the Robust Kalman Filter solution is performed in an FPGA. The calculation strategy is based on matrix factorization using Givens rotation and, in sequence, using the reverse substitution algorithm. Thus, the analysis of configurations in a heterogeneous software/hardware approach can provide possible improvements in the computational performance of the air navigation system.