Software design for dual-computer configuration of aerial robots

Abstract

Recent advances in computing have lead to low-power computers with powerful graphics processing units that can fit on common micro aerial robotics platforms. These computers carry the computational firepower to run complex algorithms, including deep learning tasks and vision-based state estimation. However, they typically lack the hardware and software for real-time control and state estimation, which is critical in aerial robotics applications. An effective solution to this problem is a dual-computer design which splits the workload between a timing-critical microcontroller and a high-performance processor. This thesis describes the implementation of certain pieces of software for this design, and the communication between the two processors.