Drone racing has recently became a topic of interest in research especially with the increase of power of mobile processors. There are many approaches of localizing (perception), planning, and strategizing against an adversarial agent online, with varying degrees of computational complexity and success. This thesis presents a game theoretic approach to solve this problem in the context of drone racing. The game theory planner strategizes against an opponent by using the “iterated best response” learning method from game theory, to attempt to reach a Nash equilibrium, where neither players can improve their strategy. Furthermore, to complement the functionality of the game theory planner, a polynomial trajectory generation algorithm is used to generate a reference track. Lastly, a model predictive controller is used to execute the strategic path generated by the game theory planner. The game theory planner performed better than the pure MPC by holding the lead position significantly longer, even though it had lower maximum velocity. On the other hand, the pure MPC held the lead position for a shorter time when the roles were switched. Furthermore, the game theory planner also performed better against the policy improvement racer. However, the policy improvement racer fared better against the game theory planner compared to the pure MPC. Lastly, some intuitive competitive behaviors such as blocking and overtaking came naturally as a result of the algorithm.
|Date of Award||Nov 2020|
|Original language||English (US)|
- Computer, Electrical and Mathematical Science and Engineering
|Supervisor||Jeff Shamma (Supervisor)|
- Game Theory
- Drone Racing