Vision-driven UAV River Following: Benchmarking with Safe Reinforcement Learning

TL;DR

This paper benchmarks Safe Reinforcement Learning algorithms for vision-based UAV river following in a realistic simulation, highlighting the effectiveness of semantic image encoding and on-policy methods for safety and performance.

Contribution

It introduces a semantic-augmented image encoding method and provides a comprehensive benchmark of Safe RL algorithms for UAV river following, demonstrating the superiority of on-policy algorithms.

Findings

First Order Constrained Optimization balances reward and safety.

On-policy algorithms outperform off-policy and model-based methods.

Semantic encoding improves water pixel reconstruction and state representation.

Abstract

In this study, we conduct a comprehensive benchmark of the Safe Reinforcement Learning (Safe RL) algorithms for the task of vision-driven river following of Unmanned Aerial Vehicle (UAV) in a Unity-based photo-realistic simulation environment. We empirically validate the effectiveness of semantic-augmented image encoding method, assessing its superiority based on Relative Entropy and the quality of water pixel reconstruction. The determination of the encoding dimension, guided by reconstruction loss, contributes to a more compact state representation, facilitating the training of Safe RL policies. Across all benchmarked Safe RL algorithms, we find that First Order Constrained Optimization in Policy Space achieves the optimal balance between reward acquisition and safety compliance. Notably, our results reveal that on-policy algorithms consistently outperform both off-policy and model-based counterparts in both training and testing environments. Importantly, the benchmarking outcomes and the vision encoding methodology extend beyond UAVs, and are applicable to Autonomous Surface Vehicles (ASVs) engaged in autonomous navigation in confined waters.