An LSTM Network for Real-Time Odometry Estimation

Valente, Michelle; Joly, Cyril; Fortelle, Arnaud de La

doi:10.1109/ivs.2019.8814133

Cited by 14 publications

(21 citation statements)

References 24 publications

(36 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We chose these three sequences because they are not very long, leaving more data for the training, but they can still be challenging and present the potential of the proposed method. In the previous work [19], we could not evaluate the sequence 01 because the vehicle is in a highway where the 2D laser scanner could not detect obstacles most of the time, making it impossible for a laser-only network to predict the odometry. For this reason, we add this sequence specially to observe if the fusion could improve the results.…”

Section: Resultsmentioning

confidence: 99%

“…In [19], we prove that the CNN network get better results if we reformulate the problem as a classification task for the rotation, and continue it as a regression one for the translation. Considering all the possible variation of angles between two frames, we created classes for the interval ±5.6 • with 0.1 • resolution, resulting in 112 possible classes.…”

Section: Trainingmentioning

confidence: 89%

“…The use of 2D laser scanners instead of 3D laser scanners could reduce considerably the price of future intelligent vehicles and the need of high computational resources. The authors of this work presented in [19] a solution that relies only on a 2D laser scanner for odometry estimation using Recurrent Convolutional Neural Networks (RCNNs). The method showed promising results along with a very fast computational time.…”

Section: Related Workmentioning

confidence: 99%

“…The raw data coming from the two sensors need to be prepared before they can be used as an input for the neural network. For the laser scanner we use the same data encoding of our previous work [19], where the sensor point set is encoded into a 1D vector. The vector is created by binning the raw scans into bins of resolution 0.1 • .…”

Section: A Data Pre-processingmentioning

confidence: 99%

“…Sequentially, the tensor is fed into the sequence of 1D convolutional and average pool layers to learn the features between the two acquisitions. We use the same CNN configuration of our previous work [19]. It consists of 6 1D convolutional layers, where each layer is followed by a rectified linear unit (ReLU) activation.…”

Section: B Network Architecturementioning

confidence: 99%

See 4 more Smart Citations

Deep Sensor Fusion for Real-Time Odometry Estimation

Valente

Joly

Fortelle

2019

2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Self Cite

View full text Add to dashboard Cite

Cameras and 2D laser scanners, in combination, are able to provide low-cost, light-weight and accurate solutions, which make their fusion well-suited for many robot navigation tasks. However, correct data fusion depends on precise calibration of the rigid body transform between the sensors. In this paper we present the first framework that makes use of Convolutional Neural Networks (CNNs) for odometry estimation fusing 2D laser scanners and mono-cameras. The use of CNNs provides the tools to not only extract the features from the two sensors, but also to fuse and match them without needing a calibration between the sensors. We transform the odometry estimation into an ordinal classification problem in order to find accurate rotation and translation values between consecutive frames. Results on a real road dataset show that the fusion network runs in real-time and is able to improve the odometry estimation of a single sensor alone by learning how to fuse two different types of data information.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Trainingmentioning

confidence: 89%

Section: Related Workmentioning

confidence: 99%

Section: A Data Pre-processingmentioning

confidence: 99%

Section: B Network Architecturementioning

confidence: 99%

See 3 more Smart Citations

Deep Sensor Fusion for Real-Time Odometry Estimation

Valente

Joly

Fortelle

2019

2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Self Cite

View full text Add to dashboard Cite

show abstract

DeepLO: Multi-projection deep LIDAR odometry for space orbital robotics rendezvous relative navigation

Kechagias‐Stamatis

Aouf

Dubanchet³

et al. 2020

Acta Astronautica

View full text Add to dashboard Cite

This work proposes a new Light Detection and Ranging (LIDAR) based navigation architecture that is appropriate for uncooperative relative robotic space navigation applications. In contrast to current solutions that exploit 3D LIDAR data, our architecture suggests a Deep Recurrent Convolutional Neural Network (DRCNN) that exploits multi-projected imagery of the acquired 3D LIDAR data. Advantages of the proposed DRCNN are; an effective feature representation facilitated by the Convolutional Neural Network module within DRCNN, a robust modelling of the navigation dynamics due to the Recurrent Neural Network incorporated in the DRCNN, and a low processing time. Our trials evaluate several current state-of-the-art space navigation methods on various simulated but credible scenarios that involve a satellite model developed by Thales Alenia Space (France). Additionally, we evaluate real satellite LIDAR data acquired in our lab. Results demonstrate that the proposed architecture, although trained solely on simulated data, is highly adaptable and is more appealing compared to current algorithms on both simulated and real LIDAR data scenarios affording better odometry accuracy at lower computational requirements.

show abstract

Lasers that learn: The interface of laser machining and machine learning

Mills

Grant-Jacob

2021

IET Optoelectronics

View full text Add to dashboard Cite

Laser machining is a highly flexible non-contact fabrication method used extensively across academia and industry. Whilst simulations based on fundamental understanding offer some insight into the processes, both the highly non-linear interactions between laser light and matter and the variety of materials involved mean that theoretical modelling is not particularly applicable to practical experimentation. However, recent breakthroughs in machine learning have resulted in neural networks that are capable of accurate and rapid modelling of laser machining at a scale, speed, and precision well beyond those of existing theoretical approaches with applications including 3-D surface visualisation and real-time error correction. A perspective at the intersection of laser machining and machine learning is presented, followed by a discussion of future milestones and challenges for this field.10 steps each, that corresponds to a total of 10^5 experimental trials. Because of the often highly non-linear relationships amongst parameters, the standard practice is generally the systematic collection of laser-machining data for all parameter combinations to identify the optimal combination. However, this process is both time-consuming and unfocussed and can take days or weeks, hence costing unnecessary effort, time, and money. Even when the optimal parameters have been determined, small changes during manufacturing, for example in laser power or beam shape, can result in final product quality that is below the required standard, again with associated costs in time and money. What is needed, therefore, is a set of modelling methodologies for identifying optimal parameters and providing real-time monitoring and error correction during manufacturing.However, the processes that describe laser machining, such as the light-matter interaction, heat conduction, phase change of material, and material removal, are particularly complex and hence challenging to model precisely and directly from a theoretical standpoint. The exact details of these processes depend on many factors, including laser parameters (e.g. wavelength, fluence, and pulse length), associated diffraction effects, and sample properties (e.g. absorption, reflectivity, melting temperature, and ablation threshold). In addition, different interaction effects become dominant as the temporal interaction length varies from continuous wave to ultrafast (i.e. femtosecondThis is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

show abstract

An LSTM Network for Real-Time Odometry Estimation

Cited by 14 publications

References 24 publications

Deep Sensor Fusion for Real-Time Odometry Estimation

Deep Sensor Fusion for Real-Time Odometry Estimation

DeepLO: Multi-projection deep LIDAR odometry for space orbital robotics rendezvous relative navigation

Lasers that learn: The interface of laser machining and machine learning

Contact Info

Product

Resources

About