-
Trajectories of vector fields asymptotic to formal invariant curves
Authors:
Olivier Le Gal,
Fernando Sanz Sánchez
Abstract:
We prove that a formal curve $Γ$ that is invariant by a $C^\infty$ vector field $ξ$ of $\mathbb{R}^m$ has a geometrical realization, as soon as the Taylor expansion of $ξ$ is not identically zero along $Γ$. This means that there is a trajectory $γ$ of $ξ$ which is asymptotic to $Γ$. This result solves a natural question proposed by Bonckaert nearly forty years ago. We also construct an invariant…
▽ More
We prove that a formal curve $Γ$ that is invariant by a $C^\infty$ vector field $ξ$ of $\mathbb{R}^m$ has a geometrical realization, as soon as the Taylor expansion of $ξ$ is not identically zero along $Γ$. This means that there is a trajectory $γ$ of $ξ$ which is asymptotic to $Γ$. This result solves a natural question proposed by Bonckaert nearly forty years ago. We also construct an invariant $C^0$ manifold $S$ in some open horn around $Γ$ which is composed entirely of trajectories asymptotic to $Γ$, and contains the germ of any such trajectory. If $ξ$ is analytic, we prove that there exists a trajectory asymptotic to $Γ$ which is, moreover, non-oscillating with respect to subanalytic sets.
△ Less
Submitted 12 November, 2023;
originally announced November 2023.
-
Learning to Explore Indoor Environments using Autonomous Micro Aerial Vehicles
Authors:
Yuezhan Tao,
Eran Iceland,
Beiming Li,
Elchanan Zwecher,
Uri Heinemann,
Avraham Cohen,
Amir Avni,
Oren Gal,
Ariel Barel,
Vijay Kumar
Abstract:
In this paper, we address the challenge of exploring unknown indoor aerial environments using autonomous aerial robots with Size Weight and Power (SWaP) constraints. The SWaP constraints induce limits on mission time requiring efficiency in exploration. We present a novel exploration framework that uses Deep Learning (DL) to predict the most likely indoor map given the previous observations, and D…
▽ More
In this paper, we address the challenge of exploring unknown indoor aerial environments using autonomous aerial robots with Size Weight and Power (SWaP) constraints. The SWaP constraints induce limits on mission time requiring efficiency in exploration. We present a novel exploration framework that uses Deep Learning (DL) to predict the most likely indoor map given the previous observations, and Deep Reinforcement Learning (DRL) for exploration, designed to run on modern SWaP constraints neural processors. The DL-based map predictor provides a prediction of the occupancy of the unseen environment while the DRL-based planner determines the best navigation goals that can be safely reached to provide the most information. The two modules are tightly coupled and run onboard allowing the vehicle to safely map an unknown environment. Extensive experimental and simulation results show that our approach surpasses state-of-the-art methods by 50-60% in efficiency, which we measure by the fraction of the explored space as a function of the length of the trajectory traveled.
△ Less
Submitted 13 September, 2023;
originally announced September 2023.
-
Agility and Target Distribution in the Dynamic Stochastic Traveling Salesman Problem
Authors:
Aviv Adler,
Oren Gal,
Sertac Karaman
Abstract:
An important variant of the classic Traveling Salesman Problem (TSP) is the Dynamic TSP, in which a system with dynamic constraints is tasked with visiting a set of n target locations (in any order) in the shortest amount of time. Such tasks arise naturally in many robotic motion planning problems, particularly in exploration, surveillance and reconnaissance, and classical TSP algorithms on graphs…
▽ More
An important variant of the classic Traveling Salesman Problem (TSP) is the Dynamic TSP, in which a system with dynamic constraints is tasked with visiting a set of n target locations (in any order) in the shortest amount of time. Such tasks arise naturally in many robotic motion planning problems, particularly in exploration, surveillance and reconnaissance, and classical TSP algorithms on graphs are typically inapplicable in this setting. An important question about such problems is: if the target points are random, what is the length of the tour (either in expectation or as a concentration bound) as n grows? This problem is the Dynamic Stochastic TSP (DSTSP), and has been studied both for specific important vehicle models and for general dynamic systems; however, in general only the order of growth is known. In this work, we explore the connection between the distribution from which the targets are drawn and the dynamics of the system, yielding a more precise lower bound on tour length as well as a matching upper bound for the case of symmetric (or driftless) systems. We then extend the symmetric dynamics results to the case when the points are selected by a (non-random) adversary whose goal is to maximize the length, thus showing worst-case bounds on the tour length.
△ Less
Submitted 31 January, 2023;
originally announced February 2023.
-
Deep Learning on Home Drone: Searching for the Optimal Architecture
Authors:
Alaa Maalouf,
Yotam Gurfinkel,
Barak Diker,
Oren Gal,
Daniela Rus,
Dan Feldman
Abstract:
We suggest the first system that runs real-time semantic segmentation via deep learning on a weak micro-computer such as the Raspberry Pi Zero v2 (whose price was \$15) attached to a toy-drone. In particular, since the Raspberry Pi weighs less than $16$ grams, and its size is half of a credit card, we could easily attach it to the common commercial DJI Tello toy-drone (<\$100, <90 grams, 98…
▽ More
We suggest the first system that runs real-time semantic segmentation via deep learning on a weak micro-computer such as the Raspberry Pi Zero v2 (whose price was \$15) attached to a toy-drone. In particular, since the Raspberry Pi weighs less than $16$ grams, and its size is half of a credit card, we could easily attach it to the common commercial DJI Tello toy-drone (<\$100, <90 grams, 98 $\times$ 92.5 $\times$ 41 mm). The result is an autonomous drone (no laptop nor human in the loop) that can detect and classify objects in real-time from a video stream of an on-board monocular RGB camera (no GPS or LIDAR sensors). The companion videos demonstrate how this Tello drone scans the lab for people (e.g. for the use of firefighters or security forces) and for an empty parking slot outside the lab.
Existing deep learning solutions are either much too slow for real-time computation on such IoT devices, or provide results of impractical quality. Our main challenge was to design a system that takes the best of all worlds among numerous combinations of networks, deep learning platforms/frameworks, compression techniques, and compression ratios. To this end, we provide an efficient searching algorithm that aims to find the optimal combination which results in the best tradeoff between the network running time and its accuracy/performance.
△ Less
Submitted 21 September, 2022;
originally announced September 2022.
-
Integrating Deep Reinforcement and Supervised Learning to Expedite Indoor Map**
Authors:
Elchanan Zwecher,
Eran Iceland,
Sean R. Levy,
Shmuel Y. Hayoun,
Oren Gal,
Ariel Barel
Abstract:
The challenge of map** indoor environments is addressed. Typical heuristic algorithms for solving the motion planning problem are frontier-based methods, that are especially effective when the environment is completely unknown. However, in cases where prior statistical data on the environment's architectonic features is available, such algorithms can be far from optimal. Furthermore, their calcu…
▽ More
The challenge of map** indoor environments is addressed. Typical heuristic algorithms for solving the motion planning problem are frontier-based methods, that are especially effective when the environment is completely unknown. However, in cases where prior statistical data on the environment's architectonic features is available, such algorithms can be far from optimal. Furthermore, their calculation time may increase substantially as more areas are exposed. In this paper we propose two means by which to overcome these shortcomings. One is the use of deep reinforcement learning to train the motion planner. The second is the inclusion of a pre-trained generative deep neural network, acting as a map predictor. Each one helps to improve the decision making through use of the learned structural statistics of the environment, and both, being realized as neural networks, ensure a constant calculation time. We show that combining the two methods can shorten the duration of the map** process by up to 4 times, compared to frontier-based motion planning.
△ Less
Submitted 27 February, 2022; v1 submitted 17 September, 2021;
originally announced September 2021.
-
Compressing Neural Networks: Towards Determining the Optimal Layer-wise Decomposition
Authors:
Lucas Liebenwein,
Alaa Maalouf,
Oren Gal,
Dan Feldman,
Daniela Rus
Abstract:
We present a novel global compression framework for deep neural networks that automatically analyzes each layer to identify the optimal per-layer compression ratio, while simultaneously achieving the desired overall compression. Our algorithm hinges on the idea of compressing each convolutional (or fully-connected) layer by slicing its channels into multiple groups and decomposing each group via l…
▽ More
We present a novel global compression framework for deep neural networks that automatically analyzes each layer to identify the optimal per-layer compression ratio, while simultaneously achieving the desired overall compression. Our algorithm hinges on the idea of compressing each convolutional (or fully-connected) layer by slicing its channels into multiple groups and decomposing each group via low-rank decomposition. At the core of our algorithm is the derivation of layer-wise error bounds from the Eckart Young Mirsky theorem. We then leverage these bounds to frame the compression problem as an optimization problem where we wish to minimize the maximum compression error across layers and propose an efficient algorithm towards a solution. Our experiments indicate that our method outperforms existing low-rank compression approaches across a wide range of networks and data sets. We believe that our results open up new avenues for future research into the global performance-size trade-offs of modern neural networks. Our code is available at https://github.com/lucaslie/torchprune.
△ Less
Submitted 18 November, 2021; v1 submitted 23 July, 2021;
originally announced July 2021.
-
Solutions of definable ODEs with regular separation and dichotomy interlacement versus Hardy
Authors:
Olivier Le Gal,
Mickaël Matusinski,
Fernando Sanz Sánchez
Abstract:
We introduce a notion of regular separation for solutions of systems of ODEs $y'=F(x,y)$, where F is definable in a polynomially bounded o-minimal structure and $y = (y_1,y_2)$. Given a pair of solutions with flat contact, we prove that, if one of them has the property of regular separation, the pair is either interlaced or generates a Hardy field. We adapt this result to trajectories of three-dim…
▽ More
We introduce a notion of regular separation for solutions of systems of ODEs $y'=F(x,y)$, where F is definable in a polynomially bounded o-minimal structure and $y = (y_1,y_2)$. Given a pair of solutions with flat contact, we prove that, if one of them has the property of regular separation, the pair is either interlaced or generates a Hardy field. We adapt this result to trajectories of three-dimensional vector fields with definable coefficients. In the particular case of real analytic vector fields, it improves the dichotomy interlaced/separated of certain integral pencils obtained by F. Cano, R. Moussu and the third author. In this context, we show that the set of trajectories with the regular separation property and asymptotic to a formal invariant curve is never empty and it is represented by a subanalytic set of minimal dimension containing the curve. Finally, we show how to construct examples of formal invariant curves which are transcendental with respect to subanalytic sets, using the so-called (SAT) property introduced by J.-P. Rolin, R. Shaefke and the third author.
△ Less
Submitted 29 January, 2022; v1 submitted 13 December, 2020;
originally announced December 2020.
-
Powers are easy to avoid
Authors:
Gareth Jones,
Olivier Le Gal
Abstract:
Suppose that $\widetilde{\mathbb R}$ is an o-minimal expansion of the real field in which restricted power functions are definable. We show that if $\widehat{\mathbb R}$ is both a reduct (in the sense of definability) of the expansion $\widetilde{\mathbb R}^{\mathbb R}$ of $\widetilde{\mathbb R}$ by all real power functions and an expansion (again in the sense of definability) of…
▽ More
Suppose that $\widetilde{\mathbb R}$ is an o-minimal expansion of the real field in which restricted power functions are definable. We show that if $\widehat{\mathbb R}$ is both a reduct (in the sense of definability) of the expansion $\widetilde{\mathbb R}^{\mathbb R}$ of $\widetilde{\mathbb R}$ by all real power functions and an expansion (again in the sense of definability) of $\widetilde{\mathbb R}$, then, provided that $\widetilde{\mathbb R}$ and $\widehat{\mathbb R}$ have the same field of exponents, they define the same sets. This can be viewed as a polynomially bounded version of an old conjecture of van den Dries and Miller.
△ Less
Submitted 20 November, 2020;
originally announced November 2020.
-
Limits of tangent spaces to definable sets
Authors:
Si Tiep Dinh,
Olivier Le Gal,
Tien Son Pham
Abstract:
We study the set of tangent limits at a given point to a set definable in any o-minimal structure by characterizing the set of exceptional rays in the tangent cone to the set at that point and investigating the set of tangent limits along these rays. Several criteria for determining exceptional rays will be given. The main results of the paper generalize, to the o-minimal setting and to arbitrary…
▽ More
We study the set of tangent limits at a given point to a set definable in any o-minimal structure by characterizing the set of exceptional rays in the tangent cone to the set at that point and investigating the set of tangent limits along these rays. Several criteria for determining exceptional rays will be given. The main results of the paper generalize, to the o-minimal setting and to arbitrary dimension, the main results of O'Shea--Wilson which deals with algebraic surfaces in $\mathbb R^3$.
△ Less
Submitted 7 October, 2020;
originally announced October 2020.
-
On local definability of holomorphic functions
Authors:
Gareth Jones,
Jonathan Kirby,
Olivier Le Gal,
Tamara Servi
Abstract:
Given a collection A of holomorphic functions, we consider how to describe all the holomorphic functions locally definable from A. The notion of local definability of holomorphic functions was introduced by Wilkie, who gave a complete description of all functions locally definable from A in the neighbourhood of a generic point. We prove that this description is no longer complete in the neighbourh…
▽ More
Given a collection A of holomorphic functions, we consider how to describe all the holomorphic functions locally definable from A. The notion of local definability of holomorphic functions was introduced by Wilkie, who gave a complete description of all functions locally definable from A in the neighbourhood of a generic point. We prove that this description is no longer complete in the neighbourhood of non-generic points. More precisely, we produce three examples of holomorphic functions which suggest that at least three new operations need to be added to Wilkie's description in order to capture local definability in its entirety. The constructions illustrate the interaction between resolution of singularities and definability in the o-minimal setting.
△ Less
Submitted 19 December, 2017;
originally announced December 2017.
-
Tangent cones and $C^1$ regularity of definable sets
Authors:
Krzysztof Kurdyka,
Olivier Le Gal,
Nhan Nguyen
Abstract:
Let $X\subset \mathbb R^n$ be a connected locally closed definable set in an o-minimal structure. We prove that the following three statements are equivalent: (i) $X$ is a $C^1$ manifold, (ii) the tangent cone and the paratangent cone of $X$ coincide at every point in $X$, (iii) for every $x \in X$, the tangent cone of $X$ at the point $x$ is a $k$-dimensional linear subspace of $\mathbb R^n$ (…
▽ More
Let $X\subset \mathbb R^n$ be a connected locally closed definable set in an o-minimal structure. We prove that the following three statements are equivalent: (i) $X$ is a $C^1$ manifold, (ii) the tangent cone and the paratangent cone of $X$ coincide at every point in $X$, (iii) for every $x \in X$, the tangent cone of $X$ at the point $x$ is a $k$-dimensional linear subspace of $\mathbb R^n$ ($k$ does not depend on $x$) varies continuously in $x$, and the density $θ(X, x) < 3/2$.
△ Less
Submitted 15 March, 2017;
originally announced March 2017.
-
Optical elastic scattering for early label-free identification of clinical pathogens
Authors:
Valentin Genuer,
Olivier Gal,
Jérémy Méteau,
Pierre Marcoux,
Emmanuelle Schultz,
Éric Lacot,
Max Maurin,
Jean-Marc Dinten
Abstract:
We report here on the ability of elastic light scattering in discriminating Gram+, Gram-and yeasts at an early stage of growth (6h). Our technique is non-invasive, low cost and does require neither skilled operators nor reagents. Therefore it is compatible with automation. It is based on the analysis of the scattering pattern (scatterogram) generated by a bacterial microcolony growing on agar, whe…
▽ More
We report here on the ability of elastic light scattering in discriminating Gram+, Gram-and yeasts at an early stage of growth (6h). Our technique is non-invasive, low cost and does require neither skilled operators nor reagents. Therefore it is compatible with automation. It is based on the analysis of the scattering pattern (scatterogram) generated by a bacterial microcolony growing on agar, when placed in the path of a laser beam. Measurements are directly performed on closed Petri dishes. The characteristic features of a given scatterogram are first computed by projecting the pattern onto the Zernike orthogonal basis. Then the obtained data are compared to a database so that machine learning can yield identification result. A 10-fold cross-validation was performed on a database over 8 species (15 strains, 1906 scatterograms), at 6h of incubation. It yielded a 94% correct classification rate between Gram+, Gram-and yeasts. Results can be improved by using a more relevant function basis for projections, such as Fourier-Bessel functions. A fully integrated instrument has been installed at the Grenoble hospital's laboratory of bacteriology and a validation campaign has been started for the early screening of SA and MRSA (Staphylococcus aureus, methicillin-resistant S. aureus) carriers. Up to now, all the published studies about elastic scattering were performed in a forward mode, which is restricted to transparent media. However, in clinical diagnostics, most of media are opaque, such as blood-supplemented agar. That is why we propose a novel scheme capable of collecting back-scattered light which provides comparable results.
△ Less
Submitted 4 October, 2016;
originally announced October 2016.
-
Trajectories in interlaced integral pencils of 3-dimensional analytic vector fields are o-minimal
Authors:
Olivier Le Gal,
Fernando Sanz,
Patrick Speissegger
Abstract:
Let X be an analytic vector field defined in a neighborhood of the origin of R^3, and let I be an analytically non-oscillatory integral pencil of X; that is, I is a maximal family of analytically non-oscillatory trajectories of X at the origin all sharing the same iterated tangents. We prove that if I is interlaced, then for any trajectory T in I, the expansion of the structure generated over the…
▽ More
Let X be an analytic vector field defined in a neighborhood of the origin of R^3, and let I be an analytically non-oscillatory integral pencil of X; that is, I is a maximal family of analytically non-oscillatory trajectories of X at the origin all sharing the same iterated tangents. We prove that if I is interlaced, then for any trajectory T in I, the expansion of the structure generated over the real field by T and all globally subanalytic sets is model-complete, o-minimal and polynomially bounded.
△ Less
Submitted 10 October, 2017; v1 submitted 8 October, 2013;
originally announced October 2013.
