Performance Comparison of Deep and Shallow Network for Quadcopter Automation

Traditional learning approaches on controlling the quadcopters or helicopter deal with the performance improvement of specific trajectory by iterative methods. However, existing schemes do not discuss the accumulation of information from training trajectories. Motivated by the generalization capability of deep learning, in this paper, we try to synthesize control for trajectories using a neural network-based model. Initially, the quadrotor is assigned to a random position with a fixed target position. To reach target location, the performance of the quadrotor is observed with an emphasis on reward values which is modulated by the change in its positions and velocities. The proposed deep network is able to perform tasks such as takeover, landing, and hover. The experimental analysis yields a better performance of deep network over a shallow network.