Advances in centerline estimation for autonomous lateral control

Cudrano, Paolo; Mentasti, Simone; Matteucci, Matteo; Bersani, Mattia; Arrigoni, Stefano; Cheli, Federico

doi:10.1109/iv47402.2020.9304729

Cited by 21 publications

(35 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Cudrano et al [21] proposed an end-to-end model that integrates sensors, a modified UNet, a path prediction (PP) algorithm using Ego's heading angle θ and lateral distance ∆ to lane center predicted from UNet, and controllers in a pipeline for center-line estimation but without showing the model architecture and its details. They tested the model in a real car driving on two different racetracks in curvedness without other cars and with a maximum speed of 54 km/h.…”

Section: Related Workmentioning

confidence: 99%

“…The system is a pipeline consisting of perception sensors, DNN, and control actuators [1,2,3] with a data flow from sensors to DNN to path planning to controllers to actuators for making driving decisions of steering, acceleration, or braking in an end-to-end, autonomous, and real-time manner [1,2,3,5]. Since Pomerleau's pioneering work in the 1980s [6], a variety of end-to-end DNNs have been proposed for various tasks in autonomous driving [5,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24]. Most of these DNNs belong to singletask learning models having single (regression or probabilistic) loss function for training the model to infer single driving task (steering angle, lead car's distance, or turning etc.)…”

Section: Introductionmentioning

confidence: 99%

“…There is another issue in end-to-end driving pipeline, namely, the integral system of DNN, path planning [4,28,29,30,31,32], and control [5,21,33,34,35,36,37] algorithms that meets safety and comfort requirements. These algorithms are generally proposed and verified separately since automotive control systems are very complex varying with vehicle types and levels of automation [1,38,39,40,41,42].…”

Section: Introductionmentioning

confidence: 99%

“…The path prediction model comprises advanced post-processing methods that are particularly suitable for deep learning to deal with complexity, scalability, and homography issues [8,28,58,59,60,61]. The control model consists of learningbased lateral [5,9,10,12,14,19,21,34,37] and longitudinal [5,19,37] controllers, where the later is based on model predictive control [33,37], i.e., it is a learning-and model-based controller [37].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

End-to-End Multi-Task Deep Learning and Model Based Control Algorithm for Autonomous Driving

Lee¹,

Liu²

2021

Preprint

View full text Add to dashboard Cite

End-to-end driving with a deep learning neural network (DNN) has become a rapidly growing paradigm of autonomous driving in industry and academia. Yet safety measures and interpretability still pose challenges to this paradigm. We propose an end-toend driving algorithm that integrates multi-task DNN, path prediction, and control models in a pipeline of data flow from sensory devices through these models to driving decisions. It provides quantitative measures to evaluate the holistic, dynamic, and real-time performance of end-to-end driving systems, and thus allows to quantify their safety and interpretability. The DNN is a modified UNet, a well known encoder-decoder neural network of semantic segmentation. It consists of one segmentation, one regression, and two classification tasks for lane segmentation, path prediction, and vehicle controls. We present three variants of the modified UNet architecture having different complexities, compare them on different tasks in four static measures for both single and multi-task (MT) architectures, and then identify the best one by two additional dynamic measures in real-time simulation. We also propose a learning-and model-based longitudinal controller using model predictive control method. With the Stanley lateral controller, our results show that MTUNet outperforms an earlier modified UNet in terms of curvature and lateral offset estimation on curvy roads at normal speed, which has been tested in a real car driving on real roads.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

End-to-End Multi-Task Deep Learning and Model Based Control Algorithm for Autonomous Driving

Lee¹,

Liu²

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Road lane detection is one of the most important processes in automatic driving [1][2][3][4][5][6]. With road lane detection systems, we can analyze the video images of violations such as lane deviation.…”

Section: Introductionmentioning

confidence: 99%

A Method for Automatic Lane Detection using a Deep Network

2022

IJE

View full text Add to dashboard Cite

Lane detection is amongst the most important operations during the automatic driving process. This process aims to detect lane lines to control the vehicle's direction in a specific lane on the road and it can be effective in preventing accidents. Besides being online, the most requirement for lane detection methods in automatic driving is their high accuracy. The use of deep learning to create fully automated systems in lane detection has been done extensively. Automated learning-based methods used for road lane detection are often of supervised type. One of the disadvantages of these methods, despite their excellent accuracy, is that they need a set of labeled data, which limits the development procedure of the lane detection system, and most importantly establishing a standard set of labeled data is very timedemanding. The recommended solution is to use appropriate learning approaches that can be used to achieve the relative accuracy of the supervised approaches and to improve their speed. It also enables us to use different datasets without a constraint label that the tagged dataset would create in the algorithm development relative to the new dataset. In this research, we present an automatic semi-supervised learning method using deep neural networks to extract the data of lane lines, using labeled (mask for detected lines) and unlabeled datasets. The results demonstrate the suitable accuracy of the adopted method according to the proposed approach and also improves its computational complexity due to the significant reduction in the number of teachable network parameters.

show abstract