Competitive drone racing using asymmetric games

This paper presents a game theoretic approach to solve the problem of drone racing. A game theory planner (GTP) strategizes against an opponent by using an iterated best response learning method from game theory. To complement the functionality of the GTP, a minimum jerk polynomial trajectory generation algorithm is used to generate a reference track. Moreover, a time-varying linear model predictive controller (MPC) is used to execute the strategic path generated by the GTP. The performance of the GTP is compared against a pure MPC, a Policy Improvement (PI) racer, and itself under different parameters. Intuitive competitive behaviors such as blocking and overtaking came naturally as a result of the algorithm. Also, interesting match-up and lead-dependent performance advantage is observed.