Deep Reinforcement Learning for Mapless Navigation of Unmanned Aerial Vehicles

Abstract

This paper presents a novel deep reinforcement learning-based system for 3D mapless navigation for Unmanned Aerial Vehicles (UAVs). Instead of using a image-based sensing approach, we propose a simple learning system that uses only a few sparse range data from a distance sensor to train a learning agent. We based our approaches on two state-of-art double critic Deep-RL models: Twin Delayed Deep Deterministic Policy Gradient (TD3) and Soft Actor-Critic (SAC). We show that our two approaches manage to outperform an approach based on the Deep Deterministic Policy Gradient (DDPG) technique and the BUG2 algorithm. Also, our new Deep-RL structure based on Recurrent Neural Networks (RNNs) outperforms the current structure used to perform mapless navigation of mobile robots. Overall, we conclude that Deep-RL approaches based on double critic with Recurrent Neural Networks (RNNs) are better suited to perform mapless navigation and obstacle avoidance of UAVs.

Depth-CUPRL: Depth-Imaged Contrastive Unsupervised Prioritized Representations in Reinforcement Learning for Mapless Navigation of Unmanned Aerial Vehicles

“…The study around mapless navigation is extensively explored using terrestrial mobile robots [3]. Autonomous navigation of aerial mobile robots using Deep-RL approaches is less frequent and mainly focused on approaches that avoid the use of visual information [6], [10] or using simplified information, without contrastive learning [7]- [9], [13].…”

“…These depth images are stored in a replay memory, and it is sampled memories in the buffer to make the Contrastive and Reinforcement Learning. methods are capable of providing an action response for continuous systems [4], [5], as well as the Deep-RL control of UAVs in simulated environments [6]. It is possible to see the application of these techniques for the mapless navigation of UAVs in models of unknown dynamics in many environments [6]- [10].…”

“…methods are capable of providing an action response for continuous systems [4], [5], as well as the Deep-RL control of UAVs in simulated environments [6]. It is possible to see the application of these techniques for the mapless navigation of UAVs in models of unknown dynamics in many environments [6]- [10]. However, one of the significant challenges of Deep-RL techniques is the limitation given by the immense need for labeled data which is difficult to collect in the real world [11].…”

Comparing Deep Reinforcement Learning Algorithms’ Ability to Safely Navigate Challenging Waters

“… Havenstrøm et al (2021) applies a curriculum learning technique with the PPO algorithm to control a 6-DOF underactuated autonomous underwater vehicle (AUV), gradually increasing the presence and severity of obstacles and disturbances during the RL training process. Grando et al (2021) develops and compares two approaches based on the Deep Deterministic Policy Gradient (DDPG) and Soft Actor-Critic (SAC) RL algorithms, respectively, to navigate a simulated quadrotor drone to a target position in 3D, including air-water medium transitions. Overall, it is thus safe to say that the existing works in the available scientific literature prove the potential of RL in path following and collision avoidance with both stationary and moving obstacles.…”

Aprendizado por Reforço Profundo para Navegação sem Mapa de um Veículo Híbrido Aéreo-Aquático

“…Até o momento, o presente trabalho possui duas contribuic ¸ões científicas. A primeira foi realizada no Simpósio Latino Americano de Robótica 2020 (IEEE LARS 2020) [Grando et al 2020], trabalho que envolveu a parte de navegac ¸ão aérea 2D e que ficou entre os 10 melhores do evento. Outro artigo científico foi publicado na Conferência Internacional de Robótica e Automac ¸ão (IEEE ICRA 2021, Qualis A1) [Grando et al 2021…”