Active Perception Applied To Unmanned Aerial Vehicles Through Deep Reinforcement Learning

TL;DR

This paper presents a system using classical image processing and simple Deep Reinforcement Learning for UAV active perception, enabling environment understanding and target recognition without complex neural networks.

Contribution

It introduces a novel approach combining classical image processing with Deep RL for UAV active perception and target tracking in complex environments.

Findings

Effective target recognition on water surfaces

Performs well without complex CNNs or contrastive learning

Handles environmental uncertainties successfully

Abstract

Unmanned Aerial Vehicles (UAV) have been standing out due to the wide range of applications in which they can be used autonomously. However, they need intelligent systems capable of providing a greater understanding of what they perceive to perform several tasks. They become more challenging in complex environments since there is a need to perceive the environment and act under environmental uncertainties to make a decision. In this context, a system that uses active perception can improve performance by seeking the best next view through the recognition of targets while displacement occurs. This work aims to contribute to the active perception of UAVs by tackling the problem of tracking and recognizing water surface structures to perform a dynamic landing. We show that our system with classical image processing techniques and a simple Deep Reinforcement Learning (Deep-RL) agent is capable of perceiving the environment and dealing with uncertainties without making the use of complex Convolutional Neural Networks (CNN) or Contrastive Learning (CL).