Increasing Physics Engine Realism Using Neural Networks

The present work proposes an approach for increasing the degree of realism in game and simulation engines used in software simulations. Using neural networks, it is aimed to simulate kinematics applied in simulations such as aeroplane flight, signal propagation and the variety of simulations in the entertainment game industry in order to generate realistic events like clutter, earthquakes and crash events. The scope of this work is to devise an approach of an efficient way to simulate vehicle physics, movements and subjecting forces as well as generating a variety of realistic events during the combination of certain factors, faults and improper engineering to provide a feeling of realism to the end user. This can be increasingly important when simulating tyres on various surfaces and suspension under different mechanical conditions. This will be achieved by using real world data to teach a neural network. The results of this approach can be implemented in game engines, training vehicle simulators and etc. which will be a subject of future work.