NavREn-Rl: Learning to fly in real environment via end-to-end deep reinforcement learning using monocular images

Anwar, Malik Aqeel; Raychowdhury, Arijit

doi:10.1109/m2vip.2018.8600838

Cited by 20 publications

(12 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We consider the task of reinforcement learning based navigational system [7]. The goal of such system is to learn a navigational policy by interacting with the underlying environment achieving userdefined goals.…”

Section: Methodology and Fault Model 31 Learning-based Navigational S...mentioning

confidence: 99%

“…It helps an agent avoid navigating in unknown environments and situations; and take safe actions as needed. Recently, end-to-end learning-based techniques [7][8][9] have demonstrated considerable potential in navigation along with specialized hardware accelerators [10,11], where the agent processes raw sensor information and uses policy trained by reinforcement learning (RL) to directly produce output actions.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Analyzing and Improving Fault Tolerance of Learning-Based Navigation Systems

Wan¹,

Anwar²,

Hsiao³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Learning-based navigation systems are widely used in autonomous applications, such as robotics, unmanned vehicles and drones. Specialized hardware accelerators have been proposed for highperformance and energy-efficiency for such navigational tasks. However, transient and permanent faults are increasing in hardware systems and can catastrophically violate tasks safety. Meanwhile, traditional redundancy-based protection methods are challenging to deploy on resource-constrained edge applications. In this paper, we experimentally evaluate the resilience of navigation systems with respect to algorithms, fault models and data types from both RL training and inference. We further propose two efficient fault mitigation techniques that achieve 2× success rate and 39% qualityof-flight improvement in learning-based navigation systems.

show abstract

Section: Methodology and Fault Model 31 Learning-based Navigational S...mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Analyzing and Improving Fault Tolerance of Learning-Based Navigation Systems

Wan¹,

Anwar²,

Hsiao³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…They used double DQN [ 19 ] which is a classical RL algorithm to conduct this study and they used the depth of the image to generate the reward. To address safety issues, they created a virtual collision environment to train the aircraft first, and then completed the training in the real environment [ 14 ].…”

Section: Related Workmentioning

confidence: 99%

“…NASA implemented a project named L2F that used a modified MiG-27 foam target drone and some sensors to conduct real-time aerodynamic modeling and to learn adaptive control [ 13 ]. In 2018, Anwar and Raychowdhury successfully made an unmanned aerial vehicle (UAV) learn to fly in a real environment via end-to-end deep reinforcement learning using monocular images [ 14 ]. Shaker and Smith presented a fast reinforcement learning algorithm for an unmanned aerial vehicle to learn how to automatically land using visual information [ 15 ].…”

Section: Introductionmentioning

confidence: 99%

Learning to Have a Civil Aircraft Take Off under Crosswind Conditions by Reinforcement Learning with Multimodal Data and Preprocessing Data

Liu

Dai

Zhao

2021

Sensors

View full text Add to dashboard Cite

Autopilot technology in the field of aviation has developed over many years. However, it is difficult for an autopilot system to autonomously operate a civil aircraft under bad weather conditions. In this paper, we present a reinforcement learning (RL) algorithm using multimodal data and preprocessing data to have a civil aircraft take off autonomously under crosswind conditions. The multimodal data include the common flight status and visual information. The preprocessing is a new design that maps some flight data by nonlinear functions based on the general flight dynamics before these data are fed into the RL model. Extensive experiments under different crosswind conditions with a professional flight simulator demonstrate that the proposed method can effectively control a civil aircraft to take off under various crosswind conditions and achieve better performance than trials without visual information or preprocessing data.

show abstract

“…For this purpose a hallway environment of an engineering building was used that contains glass walls and corridors ∼ 1.5m wide and can be seen in Fig 14. Using the baseline deep reinforcement learning algorithm in a real environment is time-consuming. Hence the approach discussed in [33] was used. Using this approach, an expert user collects a set of data-points in the real environment.…”

Section: Experimental Verification With Dji Drone In Real Environmentmentioning

confidence: 99%

Autonomous Navigation via Deep Reinforcement Learning for Resource Constraint Edge Nodes Using Transfer Learning

Anwar

Raychowdhury

2020

IEEE Access

Self Cite

View full text Add to dashboard Cite

Smart and agile drones are fast becoming ubiquitous at the edge of the cloud. The usage of these drones are constrained by their limited power and compute capability. In this paper, we present a Transfer Learning (TL) based approach to reduce on-board computation required to train a deep neural network for autonomous navigation via Deep Reinforcement Learning for a target algorithmic performance. A library of 3D realistic meta-environments is manually designed using Unreal Gaming Engine and the network is trained end-toend. These trained meta-weights are then used as initializers to the network in a test environment and fine-tuned for the last few fully connected layers. Variation in drone dynamics and environmental characteristics is carried out to show robustness of the approach. Using NVIDIA GPU profiler it was shown that the energy consumption and training latency is reduced by 3.7x and 1.8x respectively without significant degradation in the performance in terms of average distance traveled before crash i.e. Mean Safe Flight (MSF). The approach is also tested on a real environment using DJI Tello drone and similar results were reported. The code of the approach can be found on GitHub: https://github.com/aqeelanwar/Deep-Reinforcement-Learning-DJI-Tello.git. The video of the drone with proposed approach will be uploaded to YouTube.

show abstract

NavREn-Rl: Learning to fly in real environment via end-to-end deep reinforcement learning using monocular images

Cited by 20 publications

References 20 publications

Analyzing and Improving Fault Tolerance of Learning-Based Navigation Systems

Analyzing and Improving Fault Tolerance of Learning-Based Navigation Systems

Learning to Have a Civil Aircraft Take Off under Crosswind Conditions by Reinforcement Learning with Multimodal Data and Preprocessing Data

Autonomous Navigation via Deep Reinforcement Learning for Resource Constraint Edge Nodes Using Transfer Learning

Contact Info

Product

Resources

About