Learning Transferable UAV for Forest Visual Perception

Chen, Lyujie; Wang, Wufan; Zhu, Jihong

doi:10.48550/arxiv.1806.03626

Cited by 5 publications

(7 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For this autonomous driving problem, deterministic policy setting is clearly a better choice than the stochastical policy setting since this is a problem where undesired behavior might produce a catastrophic consequence almost immediately such as collision. Besides that, unlike other papers [4,5,8,9] which discretize the possible action space and then use value-based reinforcement learning algorithms which enjoy the reduction of training difficulty with the very likely sacrifice in the performance as well, we choose the continuous action space setting and implement a policybased RL algorithm.…”

Section: Proposed Methodsmentioning

confidence: 99%

“…This kind of tasks involving the utilization of deep neural networks are usually trained in an end-to-end style. Some research implements supervised learning (or sometimes called imitation learning) using annotated labels [5]. Others use reinforcement learning [4,7,8,9] through interacting with the environment and use the sampled reward as the supervision signal to indicate how good or bad is a specific action at a specific state.…”

Section: Image-based Autonomous Driving or Navigationmentioning

confidence: 99%

“…Second is to randomize the environment during training such as weather and lighting conditions, object types, positions and texture as shown in [4]. In [5], the author uses three cameras installed on the left, middle and right side of a drone and keeps the drone facing forward open space all the time, then labels the images collected by left, middle, right cameras as 'go right', 'go straight', 'go left' respectively. However, this extremely simple control policy setting directly constraints the possible performance.…”

Section: Image-based Autonomous Driving or Navigationmentioning

confidence: 99%

“…Considering the possible catastrophic effect for undesired behavior in this task, clearly deterministic policy setting is a better choice in comparison to the stochastic counterpart. Besides that, unlike the discretized possible actions in [4,5,8,9] which can be very likely to limit the possible performance, ideally, continuous policy setting is favorable for this task.…”

Section: Reinforcement Learning For Deterministic and Continuous Policymentioning

confidence: 99%

“…For autonomous driving or navigation, the problem of generating control decision given a monocular camera image and other auxiliary vehicle sensors' outputs such as vehicle velocity, acceleration etc. has aroused a lot of interest in recent years [4,5,6,7,8,9]. Due to the huge development of deep learning and reinforcement learning in recent years, most of the recent works towards this task involve these two methodologies.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Improved Reinforcement Learning through Imitation Learning Pretraining Towards Image-based Autonomous Driving

Wang

Chang

2019

2019 19th International Conference on Control, Automation and Systems (ICCAS)

View full text Add to dashboard Cite

We present a training pipeline for the autonomous driving task given the current camera image and vehicle speed as the input to produce the throttle, brake, and steering control output. The simulator Airsim's [1] convenient weather and lighting API provides a sufficient diversity during training which can be very helpful to increase the trained policy's robustness. In order to not limit the possible policy's performance, we use a continuous and deterministic control policy setting. We utilize ResNet-34 [2] as our actor and critic networks with some slight changes in the fully connected layers. Considering human's mastery of this task and the high-complexity nature of this task, we first use imitation learning to mimic the given human policy and then leverage the trained policy and its weights to the reinforcement learning phase for which we use DDPG [3]. This combination shows a considerable performance boost comparing to both pure imitation learning and pure DDPG for the autonomous driving task.

show abstract

Section: Proposed Methodsmentioning

confidence: 99%

Section: Image-based Autonomous Driving or Navigationmentioning

confidence: 99%

Section: Image-based Autonomous Driving or Navigationmentioning

confidence: 99%

Section: Reinforcement Learning For Deterministic and Continuous Policymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Improved Reinforcement Learning through Imitation Learning Pretraining Towards Image-based Autonomous Driving

Wang

Chang

2019

2019 19th International Conference on Control, Automation and Systems (ICCAS)

View full text Add to dashboard Cite

show abstract

Synthetic Data for Deep Learning

Nikolenko

2021

Springer Optimization and Its Applications

188

View full text Add to dashboard Cite

Aims and ScopeOptimization has continued to expand in all directions at an astonishing rate. New algorithmic and theoretical techniques are continually developing and the diffusion into other disciplines is proceeding at a rapid pace, with a spot light on machine learning, artificial intelligence, and quantum computing. Our knowledge of all aspects of the field has grown even more profound. At the same time, one of the most striking trends in optimization is the constantly increasing emphasis on the interdisciplinary nature of the field. Optimization has been a basic tool in areas not limited to applied mathematics, engineering, medicine, economics, computer science, operations research, and other sciences.The series Springer Optimization and Its Applications (SOIA) aims to publish state-of-the-art expository works (monographs, contributed volumes, textbooks, handbooks) that focus on theory, methods, and applications of optimization. Topics covered include, but are not limited to, nonlinear optimization, combinatorial optimization, continuous optimization, stochastic optimization, Bayesian optimization, optimal control, discrete optimization, multi-objective optimization, and more. New to the series portfolio include Works at the intersection of optimization and machine learning, artificial intelligence, and quantum computing.Volumes from this series are indexed by Web of Science, zbMATH, Mathematical Reviews, and SCOPUS.

show abstract

Learn by Observation: Imitation Learning for Drone Patrolling from Videos of A Human Navigator

Fan

Chu

Zhang

et al. 2020

2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

We present an imitation learning method for autonomous drone patrolling based only on raw videos. Different from previous methods, we propose to let the drone learn patrolling in the air by observing and imitating how a human navigator does it on the ground. The observation process enables the automatic collection and annotation of data using inter-frame geometric consistency, resulting in less manual effort and high accuracy. Then a newly designed neural network is trained based on the annotated data to predict appropriate directions and translations for the drone to patrol in a lanekeeping manner as humans. Our method allows the drone to fly at a high altitude with a broad view and low risk. It can also detect all accessible directions at crossroads and further carry out the integration of available user instructions and autonomous patrolling control commands. Extensive experiments are conducted to demonstrate the accuracy of the proposed imitating learning process as well as the reliability of the holistic system for autonomous drone navigation. The codes, datasets as well as video demonstrations are available at https://vsislab.github.io/uavpatrol.

show abstract

Learning Transferable UAV for Forest Visual Perception

Cited by 5 publications

References 9 publications

Improved Reinforcement Learning through Imitation Learning Pretraining Towards Image-based Autonomous Driving

Improved Reinforcement Learning through Imitation Learning Pretraining Towards Image-based Autonomous Driving

Synthetic Data for Deep Learning

Learn by Observation: Imitation Learning for Drone Patrolling from Videos of A Human Navigator

Contact Info

Product

Resources

About