Abstract
In this paper, we propose a parallelized optimization-based framework for autonomous and safe control of quadrotor Unmanned Aerial Vehicles (UAVs). We achieve this by designing a real-time optimal trajectory planner and a time-varying collision chance-constrained model predictive controller. We consider an obstacle with unknown dynamics in the operational space of the UAV and plan time-optimal transfer maneuvers using the shifted Chebyshev pseudospectral method. We propose a novel sigmoid function-based approximation to the conditional collision avoidance constraint of UAV trajectory segments and enable automatic differentiation for achieving real-time implementation. Given the uncertain positions of the UAV and the obstacle, we propose a time-varying probability margin for the collision avoidance constraint and design a chance-constrained model predictive controller to track the reference optimal trajectory with minimum tracking error and avoid collisions in real-time. Moreover, we parallelize the trajectory planner and the controller to address their asynchronous computational execution. The scalability and effectiveness of the proposed architecture are evaluated by performance analysis through Monte Carlo and numerical simulations. Finally, the real-time feasibility of the integrated approach is validated by indoor high-speed maneuvers and dynamic collision avoidance experiments.