Learning Minimum-Time Flight in Cluttered Environments

We tackle the problem of minimum-time flight for a quadrotor through sequence waypoints in presence obstacles while exploiting full dynamics. Early works relied on simplified dynamics or polynomial trajectory representations that did not exploit actuator potential quadrotor, and, thus, resulted suboptimal solutions. Recent can plan trajectories; yet, trajectories are executed with control methods do account obstacles. Thus, successful execution such is prone to errors due model mismatch and in-flight disturbances. To this end, we leverage deep reinforcement learning classical topological path planning train robust neural-network controllers cluttered environments. The resulting neural network controller demonstrates substantially better performance up 19\% over state-of-the-art methods. More importantly, learned policy solves simultaneously online disturbances, thus achieving much higher robustness. As such, presented method achieves 100% success rate flying policies without collision, traditional approaches achieve only 40%. proposed validated both simulation real world, speeds 42km/h accelerations 3.6g.