Deep Reinforcement Learning for drone obstacle avoidance

Unmanned Aerial Vehicles (UAVs) are increasingly deployed in autonomous mis sions across complex, cluttered environments where reliable obstacle avoidance is crit ical. Traditional navigation frameworks rely on modular pipelines—separating percep tion, mapping, planning, and control—which often suff...

Descripción completa

Detalles Bibliográficos
Autor: Loc Pham, Thanh
Tipo de recurso: tesis de maestría
Fecha de publicación:2025
País:España
Institución:Varias* (Consorci de Biblioteques Universitáries de Catalunya, Centre de Serveis Científics i Acadèmics de Catalunya)
Repositorio:Recercat. Dipósit de la Recerca de Catalunya
OAI Identifier:oai:recercat.cat:10256/28367
Acceso en línea:http://hdl.handle.net/10256/28367
https://hdl.handle.net/10256/28367
Access Level:acceso abierto
Palabra clave:DRL (Deep Reinforcement Learning)
Machine learning
Aprenentatge profund (Aprenentatge automàtic)
Vehicles aeris autònoms
Autonomous aerial vehicles
UAV (Vehicle aeri no tripulat)
Drone aircraft
Robots -- Sistemes de navegació
Robots -- Navigation systems
Obstacle avoidance
id ES_3304ab5686e8bbb6a0d0d2d39087ca61
oai_identifier_str oai:recercat.cat:10256/28367
network_acronym_str ES
network_name_str España
repository_id_str
spelling Deep Reinforcement Learning for drone obstacle avoidanceLoc Pham, ThanhDRL (Deep Reinforcement Learning)Machine learningAprenentatge profund (Aprenentatge automàtic)Vehicles aeris autònomsAutonomous aerial vehiclesUAV (Vehicle aeri no tripulat)Drone aircraftRobots -- Sistemes de navegacióRobots -- Navigation systemsObstacle avoidanceUnmanned Aerial Vehicles (UAVs) are increasingly deployed in autonomous mis sions across complex, cluttered environments where reliable obstacle avoidance is crit ical. Traditional navigation frameworks rely on modular pipelines—separating percep tion, mapping, planning, and control—which often suffer from error accumulation, high computational overhead, and poor reactivity in dynamic scenarios. To address these lim itations, this thesis investigates an end-to-end deep reinforcement learning (DRL) frame work for real-time UAV obstacle avoidance using onboard depth sensing. We compare two state-of-the-art DRL algorithms, Proximal Policy Optimization (PPO) and Twin Delayed DDPG (TD3), in a continuous control setting, evaluating their train ing dynamics and performance in diverse simulated environments. Our initial experi ments highlight key failure modes such as collisions with overhead obstacles and dead end traps, caused by the policy’s limited temporal awareness. To overcome these, we propose a neural architecture that incorporates both a pretrained ResNet8-based depth encoder and two temporal reasoning mechanisms: (1) an LSTM module for recurrent memory, and (2) a stacked buffer of recent depth observations. This temporal augmen tation allows the agent to recover from occlusions and partial observability, significantly improving navigation robustness. Trained in a curriculum-based Gym-PyBullet-Drones environment, our final memory based policy achieves a 96% success rate across randomized 3D obstacle courses and out performs EGO-Planner-v2 in both success rate and adaptability. The results demonstrate that DRL policies with temporal context can match or exceed the performance of tradi tional planning pipelines while offering greater generalization and simplicity in deploy ment.9Universitat de Girona. Institut de Recerca en Visió per Computador i RobòticaVasiljević, GoranManen, Benjamin van2025info:eu-repo/semantics/masterThesisapplication/pdfhttp://hdl.handle.net/10256/28367https://hdl.handle.net/10256/28367Erasmus Mundus Joint Master in Intelligent Field Robotic Systems (IFROS)reponame:Recercat. Dipósit de la Recerca de Catalunyainstname:Varias* (Consorci de Biblioteques Universitáries de Catalunya, Centre de Serveis Científics i Acadèmics de Catalunya)InglésAttribution-NonCommercial-NoDerivatives 4.0 Internationalhttp://creativecommons.org/licenses/by-nc-nd/4.0/info:eu-repo/semantics/openAccessoai:recercat.cat:10256/283672026-05-29T05:05:01Z
dc.title.none.fl_str_mv Deep Reinforcement Learning for drone obstacle avoidance
title Deep Reinforcement Learning for drone obstacle avoidance
spellingShingle Deep Reinforcement Learning for drone obstacle avoidance
Loc Pham, Thanh
DRL (Deep Reinforcement Learning)
Machine learning
Aprenentatge profund (Aprenentatge automàtic)
Vehicles aeris autònoms
Autonomous aerial vehicles
UAV (Vehicle aeri no tripulat)
Drone aircraft
Robots -- Sistemes de navegació
Robots -- Navigation systems
Obstacle avoidance
title_short Deep Reinforcement Learning for drone obstacle avoidance
title_full Deep Reinforcement Learning for drone obstacle avoidance
title_fullStr Deep Reinforcement Learning for drone obstacle avoidance
title_full_unstemmed Deep Reinforcement Learning for drone obstacle avoidance
title_sort Deep Reinforcement Learning for drone obstacle avoidance
dc.creator.none.fl_str_mv Loc Pham, Thanh
author Loc Pham, Thanh
author_facet Loc Pham, Thanh
author_role author
dc.contributor.none.fl_str_mv Vasiljević, Goran
Manen, Benjamin van
dc.subject.none.fl_str_mv DRL (Deep Reinforcement Learning)
Machine learning
Aprenentatge profund (Aprenentatge automàtic)
Vehicles aeris autònoms
Autonomous aerial vehicles
UAV (Vehicle aeri no tripulat)
Drone aircraft
Robots -- Sistemes de navegació
Robots -- Navigation systems
Obstacle avoidance
topic DRL (Deep Reinforcement Learning)
Machine learning
Aprenentatge profund (Aprenentatge automàtic)
Vehicles aeris autònoms
Autonomous aerial vehicles
UAV (Vehicle aeri no tripulat)
Drone aircraft
Robots -- Sistemes de navegació
Robots -- Navigation systems
Obstacle avoidance
description Unmanned Aerial Vehicles (UAVs) are increasingly deployed in autonomous mis sions across complex, cluttered environments where reliable obstacle avoidance is crit ical. Traditional navigation frameworks rely on modular pipelines—separating percep tion, mapping, planning, and control—which often suffer from error accumulation, high computational overhead, and poor reactivity in dynamic scenarios. To address these lim itations, this thesis investigates an end-to-end deep reinforcement learning (DRL) frame work for real-time UAV obstacle avoidance using onboard depth sensing. We compare two state-of-the-art DRL algorithms, Proximal Policy Optimization (PPO) and Twin Delayed DDPG (TD3), in a continuous control setting, evaluating their train ing dynamics and performance in diverse simulated environments. Our initial experi ments highlight key failure modes such as collisions with overhead obstacles and dead end traps, caused by the policy’s limited temporal awareness. To overcome these, we propose a neural architecture that incorporates both a pretrained ResNet8-based depth encoder and two temporal reasoning mechanisms: (1) an LSTM module for recurrent memory, and (2) a stacked buffer of recent depth observations. This temporal augmen tation allows the agent to recover from occlusions and partial observability, significantly improving navigation robustness. Trained in a curriculum-based Gym-PyBullet-Drones environment, our final memory based policy achieves a 96% success rate across randomized 3D obstacle courses and out performs EGO-Planner-v2 in both success rate and adaptability. The results demonstrate that DRL policies with temporal context can match or exceed the performance of tradi tional planning pipelines while offering greater generalization and simplicity in deploy ment.
publishDate 2025
dc.date.none.fl_str_mv 2025
dc.type.none.fl_str_mv info:eu-repo/semantics/masterThesis
format masterThesis
dc.identifier.none.fl_str_mv http://hdl.handle.net/10256/28367
https://hdl.handle.net/10256/28367
url http://hdl.handle.net/10256/28367
https://hdl.handle.net/10256/28367
dc.language.none.fl_str_mv Inglés
language_invalid_str_mv Inglés
dc.rights.none.fl_str_mv Attribution-NonCommercial-NoDerivatives 4.0 International
http://creativecommons.org/licenses/by-nc-nd/4.0/
info:eu-repo/semantics/openAccess
rights_invalid_str_mv Attribution-NonCommercial-NoDerivatives 4.0 International
http://creativecommons.org/licenses/by-nc-nd/4.0/
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
dc.publisher.none.fl_str_mv Universitat de Girona. Institut de Recerca en Visió per Computador i Robòtica
publisher.none.fl_str_mv Universitat de Girona. Institut de Recerca en Visió per Computador i Robòtica
dc.source.none.fl_str_mv Erasmus Mundus Joint Master in Intelligent Field Robotic Systems (IFROS)
reponame:Recercat. Dipósit de la Recerca de Catalunya
instname:Varias* (Consorci de Biblioteques Universitáries de Catalunya, Centre de Serveis Científics i Acadèmics de Catalunya)
instname_str Varias* (Consorci de Biblioteques Universitáries de Catalunya, Centre de Serveis Científics i Acadèmics de Catalunya)
reponame_str Recercat. Dipósit de la Recerca de Catalunya
collection Recercat. Dipósit de la Recerca de Catalunya
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869405719640408064
score 15,198674

Similares en LA Referencia