Publication

Abstract : A Quadcopter is a variation of aerial drone which has two rotors on one side and two rotors on the other side and with control mechanism in the central part of the quadcopter. Quadcopters are widely used because of its importance in life-threatening situations like flood-rescuing, scientific data collection, land mining and even in crowd monitoring during corona crisis etc. The proposed work is to simulate a quadcopter and make it travel in an unknown, complex environment. Complex environments can be easily navigated by the quadcopter which is modeled using reinforcement learning. The proposed work implements a Neural Network agent(Algorithm) capable of learning to fly a quadcopter by itself from source to target position against random asymmetrical obstacles. The quadcopter(actor) is controlled by four rotors and it gets feedback from the environment(critic). The work uses Deep Deterministic Policy Gradients (DDPG) algorithm to train the Neural Network. The simulation is done using physics_ sim simulator in python. The model is evaluated using different measures such as Reward, distance from the target position to the best position quadcopter has reached, variance in velocity vectors, accuracy etc.