Acceleration of Kalman filter applications with heterogeneous computing

• Grant number: 22/02094-1
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: March 01, 2023
• End date: February 29, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Ricardo Menotti
• Grantee: Gustavo de Macena Barreto
• Host Institution: Centro de Ciências Exatas e de Tecnologia (CCET). Universidade Federal de São Carlos (UFSCAR). São Carlos , SP, Brazil

Abstract

The problem of estimating the attitude and position of air vehicles is largely dealt with using techniques based on Kalman filter. In fact, the algorithm obtains good estimates when the system states are not observable, as presented in several works in the last decades. In this research project, attitude estimation will be treated using a version of the Kalman filter for nonlinear systems, known as the unscented Kalman filter. In general, the unscented Kalman filter presents better results compared to the extended Kalman filter, another variation of the Kalman filter when the model that describes the dynamics of the states or measurements of the sensors is non-linear. However, the computational cost associated with the unscented version is much higher than the extended version of the filter, especially when the state space dimension is of higher order. In this perspective, the present research project proposes a solution for the estimation of attitude states based on the oneAPI programming model, with implementations tested on multicore CPUs, GPUs and FPGAs platforms. The fusion of data from the unscented Kalman filter will be performed using data from gyroscopes, accelerometers and magnetometers, integrally arranged in 3D space, giving 9 degrees of freedom. A comparative study regarding the performance of the attitude system execution on these platforms will be carried out. With this, it is expected that it will be possible to accelerate this application, verify its suitability for the available accelerators and allow them to be later embedded in those that best suit.