Vision and control capabilities of autonomous systems with the Kria Adaptive System-on-Modules

The role of computer vision in autonomous control systems is increasingly growing. Along with this, there is a steady trend of increasing requirements for sensors performance, the speed of image processing algorithms, and the complexity of their implementation with the means of digital computing. This research explores the prospects for vision and control in a potentially dynamically changing environment using the recent system-on-module (SoM) Kria manufactured by AMD/Xilinx. Acceleration based on heterogeneous systems incorporating user reconfigurable logic is considered a promising approach to improve performance and energy efficiency due to the high level of parallelism they offer. Therefore, the structure of the Kria SoM platform and the processing pipeline are examined in this paper. Object recognition experiments have been conducted which aim to demonstrate the suitability and accessibility of adaptive computing as a good candidate that can respond to the changing demands of a wide range of robotic applications.