Q-learning Based Navigation of a Quadrotor using Non-singular Terminal Sliding Mode Control

Abstract

This paper demonstrates an hybrid methodology of quadrotor navigation and control in an environment with obstacles by combining a Q-learning strategy for navigation with a non-linear sliding mode control scheme for position and altitude control of the quadrotor. In an unknown environment, an optimal safe path is estimated using the Q-learning scheme by considering the environment as a 3D grid world. Furthermore, a non-singular terminal sliding mode control (NTSMC) is employed to navigate the quadrotor through the planned trajectories. The NTSMC that is employed for trajectory tracking ensures robustness towards bounded disturbances as well as parametric uncertainties. In addition, it ensures finite time convergence of the tracking error and avoids issues that arise due to singularities in the dynamics. The effectiveness of the proposed navigation and control scheme are validated using numerical simulations wherein a quadrotor is required to pass through a window.