Institutional Repository
Technical University of Crete
EN  |  EL



My Space

Physics-based drone racing simulator invirtual reality

Oikonomakis Michail

Full record

Year 2023
Type of Item Diploma Work
Bibliographic Citation Michail Oikonomakis, "Physics-based drone racing simulator in virtual reality", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2023
Appears in Collections


In this thesis, a virtual reality drone racing simulator with the use of avionics ispresented, designed to help humans train to fly quadcopters at high speeds. With the increase of the usage of UAVs in different fields nowadays, comes the demand in having experienced pilots that can fly the aircrafts both at high speed but also avoid obstacles.In this simulator, real physics for the movement of the quadcopters is implemented. A quadcopter has four propellers in each of the four corners. For each propeller, speed and direction of rotation are controlled in order to handle the yaw, pitch, roll and thrust.Lift and drag forces were implemented in order for the movement of the quadcopter to be realistic . Wind zones were implemented using the normal distribution algorithm to make them feel more realistic. With the usage of Virtual Reality, the user is called to pass through checkpoints in three stages while also avoid obstacles in the way. This makes the training danger free in real time scenarios, since the user can restart the simulation whenever it hits an obstacle. Each stage provides a different training experience, the firstone simulates an urban and suburban environment, the second one a coastal environment with more wind zones and the third one mostly blocks and two main colors to help ease the training. A challenging opponent in the form of AI was made using reinforcement learning algorithms. It is trained using a neural network and takes the same inputs as the player drone and provides outputs on how to handle its yaw,roll, pitch and thrust.The application contains three stages and the possibility of first or third person view while the user is moving solo or versus AI opponents. Moreover, easy and hard mode is included that adjusts how well the opponent AI is trained. The evaluation of the system focuses on noting the completion time of the stage and the number of restarts. It was evaluated if the existence of the AI opponents had any impact on the training process of the user.

Available Files