LF-VIO: A Visual-Inertial-Odometry Framework for Large Field-of-View Cameras with Negative Plane

Wang, Ze; Yang, Kailun; Shi, Hao; Li, Peng; Gao, Fei; Wang, Kaiwei

doi:10.1109/iros47612.2022.9981217

Cited by 7 publications

(4 citation statements)

References 61 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To achieve this, robots need robust perception capabilities and map-building techniques [5]. Most current approaches rely on LIDAR, cameras, GPS, and inertial sensors for obtaining environmental information [6][7][8][9][10][11][12]. Robots require efficient path planning and navigation abilities to choose the best path to reach their destination [13,14].…”

Section: Related Workmentioning

confidence: 99%

Field Robot Self-Exploration Based on Deep Reinforcement Learning and Safety Control Mechanism

Kang,

Zhang,

et al. 2024

Preprint

View full text Add to dashboard Cite

This paper aims to address the issues that arise when using deep reinforcement learning algorithms for reactive robotic navigation in farmland environments. These challenges include complex network models, a large number of parameters, significant computational requirements, and the possibility of being vulnerable to local optimal solutions. Moreover, the complexity of the farmland environment poses a challenge to the existing indoor reactive navigation methods, which struggle to meet the requirements of tasks such as crop protection and human cooperation.To address these issues, the paper proposes the Field Robot Autonomous Navigation System (FRANS), which utilizes Deep Reinforcement Learning (DRL) to enable farmland robots to autonomously explore unknown environments. The system consists of two modules: Global Navigation (GN) and Local Navigation (LN). The GN module establishes goal points and implements a secure control mechanism to manage the robot's actions. This approach helps to address problems like locally optimal solutions. The LN module uses a lightweight DRL framework to develop a locally navigated motion strategy that guides the robot toward waypoints. FRANS does not require any prior knowledge and is capable of adjusting the navigation strategy in real time based on the current environment. The map is generated in real time as the robot moves through the environment. Experimental results demonstrate that FRANS outperforms similar approaches.

show abstract

Section: Related Workmentioning

confidence: 99%

Field Robot Self-Exploration Based on Deep Reinforcement Learning and Safety Control Mechanism

Kang,

Zhang,

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…Achieving this level of independence necessitates advanced perception capabilities and sophisticated mapbuilding techniques [5]. Currently, most methodologies employ an array of technologies including LIDAR, cameras, GPS, and inertial sensors to gather critical environmental data [6][7][8][9][10][11][12]. Robots necessitate sophisticated path planning and navigation capabilities to select optimal routes to their destinations [13,14].…”

Section: Related Workmentioning

confidence: 99%

Field Robot Self-Exploration Based on Deep Reinforcement Learning and Obstacle Avoidance Strategy

Kang,

Zhang,

et al. 2024

Preprint

View full text Add to dashboard Cite

This paper addresses the challenges posed by deep reinforcement learning (DRL) algorithms when deployed for reactive robotic navigation in agricultural settings. These challenges include the complexity of network models, extensive parameterization, substantial computational demands, and susceptibility to local optima. The intricacies of farmland environments further complicate the application of conventional indoor navigation methods, which falter in tasks such as crop monitoring and interaction with humans. To overcome these obstacles, we introduce the Field Robot Autonomous Navigation System (FRANS), which employs DRL to facilitate the autonomous exploration of unknown farmlands by robots. FRANS is composed of two primary modules: Global Navigation (GN) and Local Navigation (LN). The GN module sets destination points and implements a robust control mechanism to navigate the robot, effectively mitigating issues associated with local optima. Meanwhile, the LN module leverages a streamlined DRL framework to formulate a motion strategy that directs the robot towards these waypoints. Crucially, FRANS operates without prior environmental knowledge and dynamically adapts its navigation strategy in response to real-time environmental changes. The system continuously updates the map as the robot traverses the terrain. Experimental evaluations show that FRANS surpasses competing methodologies in performance.

show abstract

“…Deep learning techniques with RS modeling [17] have higher accuracy than traditional methods, but have higher computational requirements. In traditional VIO systems, LF-VIO [18] is an optimization-based VIO framework designed for cameras with an extremely large field of view. Optimization-based approaches [12] estimate parameters by using bundle adjustment or graph optimization algorithms.…”

Section: Related Workmentioning

confidence: 99%

Point feature correction based rolling shutter modeling for EKF-based visual-inertial odometry

Zhang,

Zhang

2023

Meas. Sci. Technol.

View full text Add to dashboard Cite

The application of consumer-grade rolling-shutter (RS) camera in VIO system deserves attention. In previous works, VIO systems with RS camera usually adopt different dimensions of RS modeling to represent the camera motion. Although the complex camera motion expression improves the accuracy, adding more variables greatly increases the computation load. Therefore, considering the limitations of camera motion modeling, this paper introduces a point feature correction strategy to improve performance of the system in both speed and accuracy. The optimized camera poses and high-frequency IMU are employed to correct the point features on RS image to new pixel coordinates in global-shutter view (at the middle of RS image readout time) by epipolar transfer. At the same time, we handle the zero velocity and abnormal cases of RS modeling to improve the robustness of the system. Furthermore, the readout time of the RS camera is online self-calibrated in the stage of state augmentation and can quickly converge and stabilize to near the true value. And the readout time of the RS camera is locally observable, except in two degenerate motions.  Experimental results demonstrate that the proposed method outperforms the state-of-the-art methods on a public dataset in terms of accuracy and computational cost. We port our algorithm to android-based mobile phone and our system outputs real-time trajectories on mobile phone. Then, we use the motion capture system to obtain the ground truth phone pose to test the performance of our algorithm.

show abstract

LF-VIO: A Visual-Inertial-Odometry Framework for Large Field-of-View Cameras with Negative Plane

Cited by 7 publications

References 61 publications

Field Robot Self-Exploration Based on Deep Reinforcement Learning and Safety Control Mechanism

Field Robot Self-Exploration Based on Deep Reinforcement Learning and Safety Control Mechanism

Field Robot Self-Exploration Based on Deep Reinforcement Learning and Obstacle Avoidance Strategy

Point feature correction based rolling shutter modeling for EKF-based visual-inertial odometry

Contact Info

Product

Resources

About