Date of Award
Master of Science
Integration of reinforcement learning with unmanned aerial vehicles (UAVs) to achieve autonomous ﬂight has been an active research area in recent years. An important part focuses on collision detection and avoidance as a UAV navigates through an environment. In this thesis, we introduce a new variation of the Deep Q-Network (DQN) algorithm for UAV collision avoidance. Exploration with other variations of DQN for collision avoidance such as D3QN, are typically done through uniform sampling of actions, however, the challenge is environments inherently have sparse rewards resulting in many actions leading to redundant states. We focus on this problem of learning the dynamics of an unseen environment with sparse rewards more eﬃciently. To this end, we present an algorithm for improved exploration for UAVs. The approach is a guidance based method that uses a Bayesian Gaussian mixture model to compare previously seen states to a predicted next state in order to select the next action. Performance of these approaches was demonstrated in multiple simulation environments using Microsoft AirSim. The proposed algorithm demonstrates a two-fold improvement in average rewards compared to D3QN, after the ﬁrst 1000 training episodes.
Roghair, Jeremy, "A vision based DQN exploration algorithm for collision avoidance" (2020). Graduate Theses and Dissertations. 17984.