arXiv subjects

O. Michel

Publications and source records attributed to O. Michel.

4 recordsLinked to original sources

arXiv2022

OpenDR: An Open Toolkit for Enabling High Performance, Low Footprint Deep Learning for Robotics

Existing Deep Learning (DL) frameworks typically do not provide ready-to-use solutions for robotics, where very specific learning, reasoning, and embodiment problems exist. Their relatively steep learning curve and the different methodologies employed by DL compared to traditional approaches, along with the high complexity of DL models, which often leads to the need of employing specialized hardware accelerators, further increase the effort and cost needed to employ DL models in robotics. Also, most of the existing DL methods follow a static inference paradigm, as inherited by the traditional computer vision pipelines, ignoring active perception, which can be employed to actively interact with the environment in order to increase perception accuracy. In this paper, we present the Open Deep Learning Toolkit for Robotics (OpenDR). OpenDR aims at developing an open, non-proprietary, efficient, and modular toolkit that can be easily used by robotics companies and research institutions to efficiently develop and deploy AI and cognition technologies to robotics applications, providing a solid step towards addressing the aforementioned challenges. We also detail the design choices, along with an abstract interface that was created to overcome these challenges. This interface can describe various robotic tasks, spanning beyond traditional DL cognition and inference, as known by existing frameworks, incorporating openness, homogeneity and robotics-oriented perception e.g., through active perception, as its core design principles.

cs.RO↗

arXiv2013

New algorithm for footstep localization using seismic sensors in an indoor environment

In this study, we consider the use of seismic sensors for footstep localization in indoor environments. A popular strategy of localization is to use the measured differences in arrival times of source signals at multiple pairs of receivers. In the literature, most algorithms that are based on time differences of arrival (TDOA) assume that the propagation velocity is a constant as a function of the source position, which is valid for air propagation or even for narrow band signals. However a bounded medium such as a concrete slab (encountered in indoor environement) is usually dispersive and damped. In this study, we demonstrate that under such conditions, the concrete slab can be assimilated to a thin plate; considering a Kelvin-Voigt damping model, we introduce the notion of {\em perceived propagation velocity}, which decreases when the source-sensor distance increases. This peculiar behaviour precludes any possibility to rely on existing localization methods in indoor environment. Therefore, a new localization algorithm that is adapted to a damped and dispersive medium is proposed, using only on the sign of the measured TDOA (SO-TDOA). A simulation and some experimental results are included, to define the performance of this SO-TDOA algorithm.

cs.CE↗

arXiv2001

Measurement of Lagrangian velocity in fully developed turbulence

We have developed a new experimental technique to measure the Lagrangian velocity of tracer particles in a turbulent flow, based on ultrasonic Doppler tracking. This method yields a direct access to the velocity of a single particule at a turbulent Reynolds number $R_λ = 740$. Its dynamics is analyzed with two decades of time resolution, below the Lagrangian correlation time. We observe that the Lagrangian velocity spectrum has a Lorentzian form $E^{L}(ω) = u_{rms}^{2} T_{L} / (1 + (T_{L}ω)^{2})$, in agreement with a Kolmogorov-like scaling in the inertial range. The probability density function (PDF) of the velocity time increments displays a change of shape from quasi-Gaussian a integral time scale to stretched exponential tails at the smallest time increments. This intermittency, when measured from relative scaling exponents of structure functions, is more pronounced than in the Eulerian framework.

physics.flu-dyn↗

arXiv2001

Time resolved tracking of a sound scatterer in a turbulent flow: non-stationary signal analysis and applications

It is known that ultrasound techniques yield non-intrusive measurements of hydrodynamic flows. For example, the study of the echoes produced by a large number of particle insonified by pulsed wavetrains has led to a now standard velocimetry technique. In this paper, we propose to extend the method to the continuous tracking of one single particle embedded in a complex flow. This gives a Lagrangian measurement of the fluid motion, which is of importance in mixing and turbulence studies. The method relies on the ability to resolve in time the Doppler shift of the sound scattered by the continuously insonfied particle. For this signal processing problem two classes of approaches are used: time-frequency analysis and parametric high resolution methods. In the first class we consider the spectrogram and reassigned spectrogram, and we apply it to detect the motion of a small bead settling in a fluid at rest. In more non-stationary turbulent flows where methods in the second class are more robust, we have adapted an Approximated Maximum Likelihood technique coupled with a generalized Kalman filter.

physics.flu-dyn↗