arXiv subjects

Ovidiu Daescu

Publications and source records attributed to Ovidiu Daescu.

At least 19 recordsLinked to original sources

arXiv2026

Grow-Prune-Freeze Networks: Adaptive & Continual Learning Technique for Olfactory Navigation

Training data for olfaction is scattered through disparate, non-standardized datasets that limit the ability to build representative world models. Olfactory navigation is a highly dynamic and non-stationary task that benefits from real-time continual learning. We introduce an adaptive framework called Grow-Prune-Freeze (GPF) networks that enable an agent to continually learn through growing, pruning, and freezing early layers of its policy in response to world complexity. Grounding GPFs in non-linear random matrix theory, we show that the work of Pennington & Worth (2017) can be extended from single hidden layers to n-layer continual-learning models, and that eigenvalue composition of network weights is preserved as successive layers are added. We show that GPFs based on Expected SARSA achieve a 94% success rate on turbulent plume navigation - a partially observable, non-stationary task representative of the "big world" challenges that motivate adaptive learning in robotics - and provide supporting methodology for applying GPFs in other world models. Further experiments amount evidence that GPFs may generalize well to other machine learning tasks such as reinforcement learning in Atari, image classification, and autoregressive language models. We open source all code and data to encourage improvements on and more research in olfactory robotics.

cs.LG↗

arXiv2026

BiScale-GTR: Fragment-Aware Graph Transformers for Multi-Scale Molecular Representation Learning

Fragment-level representations provide a natural way to capture recurring molecular substructures and reuse their learned representations across molecules. However, a shared fragment identity alone may not fully describe how a fragment is instantiated in a particular molecule, since the same fragment can exhibit different chemical behavior depending on its surrounding atomic environment. Effective fragment-based molecular learning therefore requires representations that are both reusable across molecules and sensitive to local atomic context. We introduce BiScale-GTR, a self-supervised molecular representation framework built around context-grounded shared fragment tokens. BiScale-GTR constructs a reusable graph Byte Pair Encoding (graph-BPE) vocabulary using Weisfeiler-Lehman (WL)-based fragment identity, chemical validity filtering, and recursive out-of-vocabulary (OOV) decomposition. Each shared fragment token is then grounded with atom-level GNN representations through atom-to-fragment pooling and gated fusion, allowing the same fragment identity to acquire context-dependent representations in different molecular environments. A structure-aware fragment Transformer performs global reasoning over these atom-grounded tokens, capturing reusable substructure identity, local chemical context, and long-range molecular dependencies. Experiments on MoleculeNet, PharmaBench, and the Long Range Graph Benchmark demonstrate strong performance across classification and regression tasks. Attribution analysis further shows that BiScale-GTR highlights chemically meaningful recurring motifs, providing interpretable links between molecular structure and predicted properties.

cs.LG↗

arXiv2026

Chasing Ghosts: A Simulation-to-Real Olfactory Navigation Stack with Optional Vision Augmentation

Autonomous odor source localization remains a challenging problem for aerial robots due to turbulent airflow, sparse and delayed sensory signals, and strict payload and compute constraints. While prior unmanned aerial vehicle (UAV)-based olfaction systems have demonstrated gas distribution mapping or reactive plume tracing, they rely on predefined coverage patterns, external infrastructure, or extensive sensing and coordination. In this work, we present a complete, open-source UAV system for online odor source localization using a minimal sensor suite. The system integrates custom olfaction hardware, onboard sensing, and a learning-based navigation policy trained in simulation and deployed on a real quadrotor. Through our minimal framework, the UAV is able to navigate directly toward an odor source without constructing an explicit gas distribution map or relying on external positioning systems. Vision is incorporated as an optional complementary modality to accelerate navigation under certain conditions. We validate the proposed system through real-world flight experiments in a large indoor environment using an ethanol source, demonstrating consistent source-finding behavior under realistic airflow conditions. The primary contribution of this work is a reproducible system and methodological framework for UAV-based olfactory navigation and source finding under minimal sensing assumptions. We elaborate on our hardware design and open source our UAV firmware, simulation code, olfaction-vision dataset, and circuit board to the community. Code, data, and designs will be made available at https://github.com/KordelFranceTech/ChasingGhosts.

cs.RO↗

arXiv2025

Olfactory Inertial Odometry: Sensor Calibration and Drift Compensation

Visual inertial odometry (VIO) is a process for fusing visual and kinematic data to understand a machine's state in a navigation task. Olfactory inertial odometry (OIO) is an analog to VIO that fuses signals from gas sensors with inertial data to help a robot navigate by scent. Gas dynamics and environmental factors introduce disturbances into olfactory navigation tasks that can make OIO difficult to facilitate. With our work here, we define a process for calibrating a robot for OIO that generalizes to several olfaction sensor types. Our focus is specifically on calibrating OIO for centimeter-level accuracy in localizing an odor source on a slow-moving robot platform to demonstrate use cases in robotic surgery and touchless security screening. We demonstrate our process for OIO calibration on a real robotic arm and show how this calibration improves performance over a cold-start olfactory navigation task.

cs.RO↗

arXiv2025

Olfactory Inertial Odometry: Methodology for Effective Robot Navigation by Scent

Olfactory navigation is one of the most primitive mechanisms of exploration used by organisms. Navigation by machine olfaction (artificial smell) is a very difficult task to both simulate and solve. With this work, we define olfactory inertial odometry (OIO), a framework for using inertial kinematics, and fast-sampling olfaction sensors to enable navigation by scent analogous to visual inertial odometry (VIO). We establish how principles from SLAM and VIO can be extrapolated to olfaction to enable real-world robotic tasks. We demonstrate OIO with three different odour localization algorithms on a real 5-DoF robot arm over an odour-tracking scenario that resembles real applications in agriculture and food quality control. Our results indicate success in establishing a baseline framework for OIO from which other research in olfactory navigation can build, and we note performance enhancements that can be made to address more complex tasks in the future.

cs.RO↗

arXiv2025

Position: Olfaction Standardization is Essential for the Advancement of Embodied Artificial Intelligence

Despite extraordinary progress in artificial intelligence (AI), modern systems remain incomplete representations of human cognition. Vision, audition, and language have received disproportionate attention due to well-defined benchmarks, standardized datasets, and consensus-driven scientific foundations. In contrast, olfaction - a high-bandwidth, evolutionarily critical sense - has been largely overlooked. This omission presents a foundational gap in the construction of truly embodied and ethically aligned super-human intelligence. We argue that the exclusion of olfactory perception from AI architectures is not due to irrelevance but to structural challenges: unresolved scientific theories of smell, heterogeneous sensor technologies, lack of standardized olfactory datasets, absence of AI-oriented benchmarks, and difficulty in evaluating sub-perceptual signal processing. These obstacles have hindered the development of machine olfaction despite its tight coupling with memory, emotion, and contextual reasoning in biological systems. In this position paper, we assert that meaningful progress toward general and embodied intelligence requires serious investment in olfactory research by the AI community. We call for cross-disciplinary collaboration - spanning neuroscience, robotics, machine learning, and ethics - to formalize olfactory benchmarks, develop multimodal datasets, and define the sensory capabilities necessary for machines to understand, navigate, and act within human environments. Recognizing olfaction as a core modality is essential not only for scientific completeness, but for building AI systems that are ethically grounded in the full scope of the human experience.

cs.AI↗

arXiv2025

Diffusion Graph Neural Networks and Dataset for Robust Olfactory Navigation in Hazard Robotics

Navigation by scent is a capability in robotic systems that is rising in demand. However, current methods often suffer from ambiguities, particularly when robots misattribute odours to incorrect objects due to limitations in olfactory datasets and sensor resolutions. To address challenges in olfactory navigation, we introduce a multimodal olfaction dataset along with a novel machine learning method using diffusion-based molecular generation that can be used by itself or with automated olfactory dataset construction pipelines. This generative process of our diffusion model expands the chemical space beyond the limitations of both current olfactory datasets and training methods, enabling the identification of potential odourant molecules not previously documented. The generated molecules can then be more accurately validated using advanced olfactory sensors, enabling them to detect more compounds and inform better hardware design. By integrating visual analysis, language processing, and molecular generation, our framework enhances the ability of olfaction-vision models on robots to accurately associate odours with their correct sources, thereby improving navigation and decision-making through better sensor selection for a target compound in critical applications such as explosives detection, narcotics screening, and search and rescue. Our methodology represents a foundational advancement in the field of artificial olfaction, offering a scalable solution to challenges posed by limited olfactory data and sensor ambiguities. Code, models, and data are made available to the community at: https://huggingface.co/datasets/kordelfrance/olfaction-vision-language-dataset.

cs.RO↗

arXiv2022

Approximating the discrete and continuous median line segments in $d$ dimensions

Consider a set $P$ of $n$ points in $\mathbb{R}^d$. In the discrete median line segment problem, the objective is to find a line segment bounded by a pair of points in $P$ such that the sum of the Euclidean distances from $P$ to the line segment is minimized. In the continuous median line segment problem, a real number $\ell>0$ is given, and the goal is to locate a line segment of length $\ell$ in $\mathbb{R}^d$ such that the sum of the Euclidean distances between $P$ and the line segment is minimized. We show how to compute $(1+εΔ)$- and $(1+ε)$-approximations to a discrete median line segment in time $O(nε^{-2d}\log n)$ and $O(n^2ε^{-d})$, respectively, where $Δ$ is the spread of line segments spanned by pairs of points. While developing our algorithms, by using the principle of pair decomposition, we derive new data structures that allow us to quickly approximate the sum of the distances from a set of points to a given line segment or point. To our knowledge, our utilization of pair decompositions for solving minsum facility location problems is the first of its kind; it is versatile and easily implementable. We prove that it is impossible to construct a continuous median line segment for $n\geq3$ non-collinear points in the plane by using only ruler and compass. In view of this, we present an $O(n^dε^{-d})$-time algorithm for approximating a continuous median line segment in $\mathbb{R}^d$ within a factor of $1+ε$. The algorithm is based upon generalizing the point-segment pair decomposition from the discrete to the continuous domain. Last but not least, we give an $(1+ε)$-approximation algorithm, whose time complexity is sub-quadratic in $n$, for solving the constrained median line segment problem in $\mathbb{R}^2$ where an endpoint or the slope of the median line segment is given at input.

cs.CG↗

arXiv2020

Characterization and Computation of Feasible Trajectories for an Articulated Probe with a Variable-Length End Segment

An articulated probe is modeled in the plane as two line segments, $ab$ and $bc$, joined at $b$, with $ab$ being very long, and $bc$ of some small length $r$. We investigate a trajectory planning problem involving the articulated two-segment probe where the length $r$ of $bc$ can be customized. Consider a set $P$ of simple polygonal obstacles with a total of $n$ vertices, a target point $t$ located in the free space such that $t$ cannot see to infinity, and a circle $S$ centered at $t$ enclosing $P$. The probe initially resides outside $S$, with $ab$ and $bc$ being collinear, and is restricted to the following sequence of moves: a straight line insertion of $abc$ into $S$ followed by a rotation of $bc$ around $b$. The goal is to compute a feasible obstacle-avoiding trajectory for the probe so that, after the sequence of moves, $c$ coincides with $t$. We prove that, for $n$ line segment obstacles, the smallest length $r$ for which there exists a feasible probe trajectory can be found in $O(n^{2+ε})$ time using $O(n^{2+ε})$ space, for any constant $ε> 0$. Furthermore, we prove that all values $r$ for which a feasible probe trajectory exists form $O(n^2)$ intervals, and can be computed in $O(n^{5/2})$ time using $O(n^{2+ε})$ space. We also show that, for a given $r$, the feasible trajectory space of the articulated probe can be characterized by a simple arrangement of complexity $O(n^2)$, which can be constructed in $O(n^2)$ time. To obtain our solutions, we design efficient data structures for a number of interesting variants of geometric intersection and emptiness query problems.

cs.CG↗

arXiv2020

Computing Feasible Trajectories for an Articulated Probe in Three Dimensions

Consider an input consisting of a set of $n$ disjoint triangular obstacles in $\mathbb{R}^3$ and a target point $t$ in the free space, all enclosed by a large sphere $S$ of radius $R$ centered at $t$. An articulated probe is modeled as two line segments $ab$ and $bc$ connected at point $b$. The length of $ab$ can be equal to or greater than $R$, while $bc$ is of a given length $r \leq R$. The probe is initially located outside $S$, assuming an unarticulated configuration, in which $ab$ and $bc$ are collinear and $b \in ac$. The goal is to find a feasible (obstacle-avoiding) probe trajectory to reach $t$, with the condition that the probe is constrained by the following sequence of moves -- a straight-line insertion of the unarticulated probe into $S$, possibly followed by a rotation of $bc$ at $b$ for at most $π/2$ radians, so that $c$ coincides with $t$. We prove that if there exists a feasible probe trajectory, then a set of extremal feasible trajectories must be present. Through careful case analysis, we show that these extremal trajectories can be represented by $O(n^4)$ combinatorial events. We present a solution approach that enumerates and verifies these combinatorial events for feasibility in overall $O(n^{4+ε})$ time using $O(n^{4+ε})$ space, for any constant $ε> 0$. The enumeration algorithm is highly parallel, considering that each combinatorial event can be generated and verified for feasibility independently of the others. In the process of deriving our solution, we design the first data structure for addressing a special instance of circular sector emptiness queries among polyhedral obstacles in three dimensional space, and provide a simplified data structure for the corresponding emptiness query problem in two dimensions.

cs.CG↗

arXiv2020

City Guarding with Limited Field of View

Drones and other small unmanned aerial vehicles are starting to get permission to fly within city limits. While video cameras are easily available in most cities, their purpose is to guard the streets at ground level. Guarding the aerial space of a city with video cameras is a problem that so far has been largely ignored. In this paper, we present bounds on the number of cameras needed to guard the city's aerial space (roofs, walls, and ground) using cameras with 180-degree range of vision (the region in front of the guard), which is common for most commercial cameras. We assume all buildings are vertical and have a rectangular base. Each camera is placed at a top corner of a building. We considered the following two versions: (i) buildings have an axis-aligned ground base and, (ii) buildings have an arbitrary orientation. We give necessary and sufficient results for (i), necessary results for (ii), and conjecture sufficiency results for (ii). Specifically, for (i) we prove a sufficiency bound of 2k + k/4 +4 on the number of vertex guards, while for (ii) we show that 3k + 1 vertex guards are sometimes necessary, where k is the total number of buildings in the city.

cs.CG↗

arXiv2019

Altitude Terrain Guarding and Guarding Uni-Monotone Polygons

We present an optimal, linear-time algorithm for the following version of terrain guarding: given a 1.5D terrain and a horizontal line, place the minimum number of guards on the line to see all of the terrain. We prove that the cardinality of the minimum guard set coincides with the cardinality of a maximum number of ``witnesses'', i.e., terrain points, no two of which can be seen by a single guard. We show that our results also apply to the Art Gallery problem in ``monotone mountains'', i.e., $x$-monotone polygons with a single edge as one of the boundary chains. This means that any monotone mountain is ``perfect'' (its guarding number is the same as its witness number); we thus establish the first non-trivial class of perfect polygons.

cs.CG↗

arXiv2018

Edge Disjoint Spanning Trees in an Undirected Graph with E=2(V-1)

Given a connected undirected graph G = [V; E] where |E| =2(|V| -1), we present two algorithms to check if G can be decomposed into two edge disjoint spanning trees, and provide such a decomposition when it exists. Unlike previous algorithms for finding edge disjoint spanning trees in general undirected graphs, based on matroids and complex in description, our algorithms are based on simple graph reduction techniques and thus easy to describe and implement. Moreover, the running time for our solutions is asymptotically faster. Specifically, ours are the first algorithms to achieve a running time that is a polylog factor from linear, approaching the 1974 linear time algorithm of Robert E. Tarjan for directed graphs. A direct implication of our result is that minimally rigid graphs, also called Laman graphs, can be recognized in almost linear time, thus answering a long standing open problem.

cs.DS↗

arXiv2018

Does a robot path have clearance c?

Most path planning problems among polygonal obstacles ask to find a path that avoids the obstacles and is optimal with respect to some measure or a combination of measures, for example an $u$-to-$v$ shortest path of clearance at least $c$, where $u$ and $v$ are points in the free space and $c$ is a positive constant. In practical applications, such as emergency interventions/evacuations and medical treatment planning, a number of $u$-to-$v$ paths are suggested by experts and the question is whether such paths satisfy specific requirements, such as a given clearance from the obstacles. We address the following path query problem: Given a set $S$ of $m$ disjoint simple polygons in the plane, with a total of $n$ vertices, preprocess them so that for a query consisting of a positive constant $c$ and a simple polygonal path $π$ with $k$ vertices, from a point $u$ to a point $v$ in free space, where $k$ is much smaller than $n$, one can quickly decide whether $π$ has clearance at least $c$ (that is, there is no polygonal obstacle within distance $c$ of $π$). To do so, we show how to solve the following related problem: Given a set $S$ of $m$ simple polygons in $\Re^{2}$, preprocess $S$ into a data structure so that the polygon in $S$ closest to a query line segment $s$ can be reported quickly. We present an $O(t \log n)$ time, $O(t)$ space preprocessing, $O((n / \sqrt{t}) \log ^{7/2} n)$ query time solution for this problem, for any $n ^{1 + ε} \leq t \leq n^{2}$. For a path with $k$ segments, this results in $O((n k / \sqrt{t}) \log ^{7/2} n)$ query time, which is a significant improvement over algorithms that can be derived from existing computational geometry methods when $k$ is small.

cs.CG↗

arXiv2018

k-Maximum Subarrays for Small k: Divide-and-Conquer made simpler

Given an array A of n real numbers, the maximum subarray problem is to find a contiguous subarray which has the largest sum. The k-maximum subarrays problem is to find k such subarrays with the largest sums. For the 1-maximum subarray the well known divide-and-conquer algorithm, presented in most textbooks, although suboptimal, is easy to implement and can be made optimal with a simple change that speeds up the combine phase. On the other hand, the only known divide-and-conquer algorithm for k > 1, that is efficient for small values of k, is difficult to implement, due to the intricacies of the combine phase. In this paper we give a divide- and-conquer solution for the k-maximum subarray problem that simplifies the combine phase considerably while preserving the overall running time. In the process of designing the combine phase of the algorithm we provide a simple, sublinear, O($k^{1/2} log^3 k$) time algorithm, for finding the k largest sums of X + Y, where X and Y are sorted arrays of size n and $k <= n^2$. The k largest sums are implicitly represented, and can be enumerated with an additional O(k) time. To our knowledge, this is the first sublinear time algorithm for this well studied problem. Unlike previous solutions, that are fairly complicated and sometimes difficult to implement, ours rely on simple operations such as merging sorted arrays, binary search, and selecting the $k^{th}$ smallest number in an array. We have implemented our algorithms and report excellent performance on test data.

cs.DS↗

arXiv2017

Maximum Area Rectangle Separating Red and Blue Points

Given a set R of n red points and a set B of m blue points, we study the problem of finding a rectangle that contains all the red points, the minimum number of blue points and has the largest area. We call such rectangle a maximum separating rectangle. We address the planar, axis-aligned (2D) version, and present an O(mlogm+n) time, O(m+n) space algorithm. The running time reduces to O(m + n) if the points are pre-sorted by one of the coordinates. We further prove that our algorithm is optimal in the decision model of computation.

cs.CG↗

arXiv2016

Approximation Algorithms for the Maximum Profit Pick-up Problem with Time Windows and Capacity Constraint

In this paper, we study the Maximum Profit Pick-up Problem with Time Windows and Capacity Constraint (MP-PPTWC). Our main results are 3 polynomial time algorithms, all having constant approximation factors. The first algorithm has an approximation ratio of $~46 (1 + (71/60 + \fracα{\sqrt{10+p}}) ε) \log T$, where: (i) $ε> 0$ and $T$ are constants; (ii) The maximum quantity supplied is $q_{max} = O(n^p) q_{min}$, for some $p > 0$, where $q_{min}$ is the minimum quantity supplied; (iii) $α> 0$ is a constant such that the optimal number of vehicles is always at least $\sqrt{10 + p} / α$. The second algorithm has an approximation ratio of $\simeq 46 (1 + ε+ \frac{(2 + α) ε}{\sqrt{10 + p}}) \log T$. Finally, the third algorithm has an approximation ratio of $\simeq 11 (1 + 2 ε) \log T$. While our algorithms may seem to have quite high approximation ratios, in practice they work well and, in the majority of cases, the profit obtained is at least 1/2 of the optimum.

cs.DS↗

arXiv2010

Minimum Sum Dipolar Spanning Tree in R^3

In this paper we consider finding a geometric minimum-sum dipolar spanning tree in R^3, and present an algorithm that takes O(n^2 log^2 n) time using O(n^2) space, thus almost matching the best known results for the planar case. Our solution uses an interesting result related to the complexity of the common intersection of n balls in R^3, of possible different radii, that are all tangent to a given point p. The problem has applications in communication networks, when the goal is to minimize the distance between two hubs or servers as well as the distance from any node in the network to the closer of the two hubs. The approach used in this paper also provides a solution to the discrete 2-center problem in R^3 within the same time and space bounds.

cs.CG↗