Automated Simulation Environment for Enhanced Radar Technology Learning Creating Scenarios using Deep Machine Learning and Stochastic Target Trajectories for Future Radar Simulation

Abstract

Abstract Simulations contribute to reduced resource requirements, faster testing of new equipment, and an often overlooked advantage: enhanced learning through interactive scenarios, immediate feedback, and practical experience in a safe environment. Whilethe conservation of resources contributes to sustainability, simulations also provide a quicker and more visual way to learn, which is particularly beneficial in complex fields such as radar technology. This accelerated learning process can expedite new employee onboarding, enabling more time to be dedicated to critical tasks where most needed. An automated simulation environment that generates new scenarios each use would further enhance these benefits. This study explores the development of such a system and how to implement it in the radar field. The work consisted of three main parts: creating a 3D environment based on real-world locations using deep machine learning, implementing automated target spawning within the 3D environment, and designing a simulation to demonstrate radar concepts for educational purposes. The creation of the environment requires a terrain map to predict height map and terrain type, achieved using semantic segmentation and a U-Net structure. The data from the network includes information about the terrain’s elevation and the corresponding surface color based on the terrain type. This data serves as the basis for creating a 3D environment using a Python 3D plotting library. Targets are created based on the environment, considering constraints such as maximum velocity, turn rate, and collision avoidance. Combining all the parts was a substantial aspect of the work. Before the system can function solely by inputting a terrain map, further improvements are necessary in the form of automatic data transfer and further dynamics implemented for the targets. Currently, the transfers between the main parts still need to be done manually. The goal is for the system to operate as an automated chain where one part feeds the next part with the necessary information. In summary, the idea of developing a fully automated simulation has been demonstrated to be effective through the tests conducted in this thesis. While the results of the project can support future research in the field, still many challenges await, such as ensuring that all parts of the simulation can automatically transfer data.