Synthetic dataset towards autonomous personal aerial vehicles

Abstract

Personal aerial vehicle (PAV) is an aircraft designed for transporting one or two persons much like a small personal car but flying in air based on Bernoulli Principle and Newton’s laws. In the coming decades, PAVs are anticipated to revolutionize urban transportation, offering a convenient and efficient mode of travel. Due to their high velocities and dense aerial traffic, manual piloting of PAVs is impractical, necessitating the development of Autonomous PAVs, which requires minimal to zero piloting skills. The safety and reliability of PAVs heavily rely on the implementation of advanced Autonomous Navigation and Control Systems (ANCS). These ANCS monitor the environment, make real-time decisions using sensor data and sophisticated algorithms, and enable the detection and avoidance of obstacles (including other PAVs) while maintaining their intended trajectory. To train ANCS based on machine learning algorithms, high-quality datasets that accurately represent realworld environments are required. Simulation-driven development (SDD) approaches provide an effective means to generate such datasets, surpassing the limitations of data collected from the physical world. Synthetic datasets offer several advantages, including higher quality and a cost-effective and efficient development process for ANCS. Higher quality of dataset reduces the risk of the ANCS performing poorly when deployed in the real world. The use of machine learning in ANCS enables accurate object detection and classification, enhancing the system’s decision-making capabilities. SDD further reduces the infrastructure requirements associated with PAV development and facilitates the safe integration of PAVs into urban settings.