Skip to main navigation Skip to search Skip to main content

PPO-NODE: End-to-End Vision-Based Obstacle Avoidance for UAVs in Unknown Complex Environments

  • Harbin Institute of Technology
  • China North Industries Group Corporation Limited

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

To address the challenge of autonomous obstacle avoidance for unmanned aerial vehicles (UAVs) in complex environments, we propose PPO-NODE, an end-to-end deep reinforcement learning method that integrates Proximal Policy Optimization (PPO) with Neural Ordinary Differential Equations (NODE). To predict the motion of dynamic obstacles, NODE is incorporated into the feature extraction process to capture continuous temporal features from sequences of depth images. Experimental evaluations conducted in a PyBullet-based 3D simulation environment demonstrate that the proposed PPO-NODE algorithm achieves a collision rate of 11.3 % in complex environments, significantly reducing the collision probability for UAVs. This effectively resolves the critical issues of prior map dependency and delayed response to dynamic obstacles inherent in traditional approaches.

Original languageEnglish
Title of host publicationProceedings of the 5th Conference on Fully Actuated System Theory and Applications, FASTA 2026
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages555-560
Number of pages6
ISBN (Electronic)9798319547323
DOIs
StatePublished - 2026
Event5th Conference on Fully Actuated System Theory and Applications, FASTA 2026 - Qinhuangdao, China
Duration: 22 May 202624 May 2026

Publication series

NameProceedings of the 5th Conference on Fully Actuated System Theory and Applications, FASTA 2026

Conference

Conference5th Conference on Fully Actuated System Theory and Applications, FASTA 2026
Country/TerritoryChina
CityQinhuangdao
Period22/05/2624/05/26

Keywords

  • collision avoidance
  • neural ordinary differential equations
  • reinforcement learning

Fingerprint

Dive into the research topics of 'PPO-NODE: End-to-End Vision-Based Obstacle Avoidance for UAVs in Unknown Complex Environments'. Together they form a unique fingerprint.

Cite this