arXiv subjects

Christian Wolf

Publications and source records attributed to Christian Wolf.

At least 55 records · Page 3Linked to original sources

arXiv2024

Multi-Object Navigation in real environments using hybrid policies

Navigation has been classically solved in robotics through the combination of SLAM and planning. More recently, beyond waypoint planning, problems involving significant components of (visual) high-level reasoning have been explored in simulated environments, mostly addressed with large-scale machine learning, in particular RL, offline-RL or imitation learning. These methods require the agent to learn various skills like local planning, mapping objects and querying the learned spatial representations. In contrast to simpler tasks like waypoint planning (PointGoal), for these more complex tasks the current state-of-the-art models have been thoroughly evaluated in simulation but, to our best knowledge, not yet in real environments. In this work we focus on sim2real transfer. We target the challenging Multi-Object Navigation (Multi-ON) task and port it to a physical environment containing real replicas of the originally virtual Multi-ON objects. We introduce a hybrid navigation method, which decomposes the problem into two different skills: (1) waypoint navigation is addressed with classical SLAM combined with a symbolic planner, whereas (2) exploration, semantic mapping and goal retrieval are dealt with deep neural networks trained with a combination of supervised learning and RL. We show the advantages of this approach compared to end-to-end methods both in simulation and a real environment and outperform the SOTA for this task.

cs.RO↗

arXiv2023

AutoNeRF: Training Implicit Scene Representations with Autonomous Agents

Implicit representations such as Neural Radiance Fields (NeRF) have been shown to be very effective at novel view synthesis. However, these models typically require manual and careful human data collection for training. In this paper, we present AutoNeRF, a method to collect data required to train NeRFs using autonomous embodied agents. Our method allows an agent to explore an unseen environment efficiently and use the experience to build an implicit map representation autonomously. We compare the impact of different exploration strategies including handcrafted frontier-based exploration, end-to-end and modular approaches composed of trained high-level planners and classical low-level path followers. We train these models with different reward functions tailored to this problem and evaluate the quality of the learned representations on four different downstream tasks: classical viewpoint rendering, map reconstruction, planning, and pose refinement. Empirical results show that NeRFs can be trained on actively collected data using just a single episode of experience in an unseen environment, and can be used for several downstream robotic tasks, and that modular trained exploration models outperform other classical and end-to-end baselines. Finally, we show that AutoNeRF can reconstruct large-scale scenes, and is thus a useful tool to perform scene-specific adaptation as the produced 3D environment models can be loaded into a simulator to fine-tune a policy of interest.

cs.CV↗

arXiv2023

XQz5: A New Ultraluminous z$\sim$5 Quasar Legacy Sample

Bright quasar samples at high redshift are useful for investigating active galactic nuclei evolution. In this study, we describe XQz5, a sample of 83 ultraluminous quasars in the redshift range $4.5 < z < 5.3$ with optical and near-infrared spectroscopic observations, with unprecendented completeness at the bright end of the quasar luminosity function. The sample is observed with the Southern Astrophysical Research Telescope, the Very Large Telescope, and the ANU 2.3m Telescope, resulting in a high-quality, moderate-resolution spectral atlas of the brightest known quasars within the redshift range. We use established virial mass relations to derive the black hole masses by measuring the observed Mg\,\textsc{ii}$λ$2799Å emission-line and we estimate the bolometric luminosity with bolometric corrections to the UV continuum. Comparisons to literature samples show that XQz5 bridges the redshift gap between other X-shooter quasar samples, XQ-100 and XQR-30, and is a brighter sample than both. Luminosity-matched lower-redshift samples host more massive black holes, which indicate that quasars at high redshift are more active than their counterparts at lower-redshift, in concordance with recent literature.

astro-ph.GA↗

arXiv2023

Computability in Dynamical Systems

In this paper we present an introduction to the area of computability in dynamical systems. This is a fairly new field which has received quite some attention in recent years. One of the central questions in this area is if relevant dynamical objects can be algorithmically presented by a Turing machine. After providing an overview of the relevant objects we discuss recent results concerning the computability of the entropy for symbolic systems and the computability of Julia sets as well as their Brolin-Lyubich measures.

math.DS↗

arXiv2023

Virial Black Hole Mass Estimates of Quasars in the XQ-100 Legacy Survey

The black hole (BH) mass and luminosity are key factors in determining how a quasar interacts with its environment. In this study, we utilise data from the European Southern Observatory Large Programme XQ-100, a high-quality sample of 100 X-shooter spectra of the most luminous quasars in the redshift range $3.5 < z < 4.5$, and measure the properties of three prominent optical and ultraviolet broad emission-lines present in the wide wavelength coverage of X-shooter: CIV, MgII, and H$β$. The line properties of all three broad lines are used for virial estimates of the BH mass and their resulting mass estimates for this sample are tightly correlated. The BH mass range is $\log{(\rm{M_{BH}}/\rm{M_\odot})} = 8.6-10.3$ with bolometric luminosities estimated from the 3000A continuum in the range $\log{(\rm{L_{bol}}/\rm{erg\,s^{-1}})} = 46.7-48.0$. Robustly determined properties of these quasars enable a variety of follow-up research in quasar astrophysics, from chemical abundance and evolution in the broad-line region to radiatively driven quasar outflows.

astro-ph.GA↗

arXiv2023

Learning with a Mole: Transferable latent spatial representations for navigation without reconstruction

Agents navigating in 3D environments require some form of memory, which should hold a compact and actionable representation of the history of observations useful for decision taking and planning. In most end-to-end learning approaches the representation is latent and usually does not have a clearly defined interpretation, whereas classical robotics addresses this with scene reconstruction resulting in some form of map, usually estimated with geometry and sensor models and/or learning. In this work we propose to learn an actionable representation of the scene independently of the targeted downstream task and without explicitly optimizing reconstruction. The learned representation is optimized by a blind auxiliary agent trained to navigate with it on multiple short sub episodes branching out from a waypoint and, most importantly, without any direct visual observation. We argue and show that the blindness property is important and forces the (trained) latent representation to be the only means for planning. With probing experiments we show that the learned representation optimizes navigability and not reconstruction. On downstream tasks we show that it is robust to changes in distribution, in particular the sim2real gap, which we evaluate with a real physical robot in a real office building, significantly improving performance.

cs.RO↗

arXiv2023

End-to-End (Instance)-Image Goal Navigation through Correspondence as an Emergent Phenomenon

Most recent work in goal oriented visual navigation resorts to large-scale machine learning in simulated environments. The main challenge lies in learning compact representations generalizable to unseen environments and in learning high-capacity perception modules capable of reasoning on high-dimensional input. The latter is particularly difficult when the goal is not given as a category ("ObjectNav") but as an exemplar image ("ImageNav"), as the perception module needs to learn a comparison strategy requiring to solve an underlying visual correspondence problem. This has been shown to be difficult from reward alone or with standard auxiliary tasks. We address this problem through a sequence of two pretext tasks, which serve as a prior for what we argue is one of the main bottleneck in perception, extremely wide-baseline relative pose estimation and visibility prediction in complex scenes. The first pretext task, cross-view completion is a proxy for the underlying visual correspondence problem, while the second task addresses goal detection and finding directly. We propose a new dual encoder with a large-capacity binocular ViT model and show that correspondence solutions naturally emerge from the training signals. Experiments show significant improvements and SOTA performance on the two benchmarks, ImageNav and the Instance-ImageNav variant, where camera intrinsics and height differ between observation and goal.

cs.CV↗

arXiv2023

Multi-Object Navigation with dynamically learned neural implicit representations

Understanding and mapping a new environment are core abilities of any autonomously navigating agent. While classical robotics usually estimates maps in a stand-alone manner with SLAM variants, which maintain a topological or metric representation, end-to-end learning of navigation keeps some form of memory in a neural network. Networks are typically imbued with inductive biases, which can range from vectorial representations to birds-eye metric tensors or topological structures. In this work, we propose to structure neural networks with two neural implicit representations, which are learned dynamically during each episode and map the content of the scene: (i) the Semantic Finder predicts the position of a previously seen queried object; (ii) the Occupancy and Exploration Implicit Representation encapsulates information about explored area and obstacles, and is queried with a novel global read mechanism which directly maps from function space to a usable embedding space. Both representations are leveraged by an agent trained with Reinforcement Learning (RL) and learned online during each episode. We evaluate the agent on Multi-Object Navigation and show the high impact of using neural implicit representations as a memory source.

cs.CV↗

arXiv2023

Learning whom to trust in navigation: dynamically switching between classical and neural planning

Navigation of terrestrial robots is typically addressed either with localization and mapping (SLAM) followed by classical planning on the dynamically created maps, or by machine learning (ML), often through end-to-end training with reinforcement learning (RL) or imitation learning (IL). Recently, modular designs have achieved promising results, and hybrid algorithms that combine ML with classical planning have been proposed. Existing methods implement these combinations with hand-crafted functions, which cannot fully exploit the complementary nature of the policies and the complex regularities between scene structure and planning performance. Our work builds on the hypothesis that the strengths and weaknesses of neural planners and classical planners follow some regularities, which can be learned from training data, in particular from interactions. This is grounded on the assumption that, both, trained planners and the mapping algorithms underlying classical planning are subject to failure cases depending on the semantics of the scene and that this dependence is learnable: for instance, certain areas, objects or scene structures can be reconstructed easier than others. We propose a hierarchical method composed of a high-level planner dynamically switching between a classical and a neural planner. We fully train all neural policies in simulation and evaluate the method in both simulation and real experiments with a LoCoBot robot, showing significant gains in performance, in particular in the real environment. We also qualitatively conjecture on the nature of data regularities exploited by the high-level planner.

cs.RO↗

arXiv2023

Beyond spectroscopy. II. Stellar parameters for over twenty million stars in the northern sky from SAGES DR1 and Gaia DR3

We present precise photometric estimates of stellar parameters, including effective temperature, metallicity, luminosity classification, distance, and stellar age, for nearly 26 million stars using the methodology developed in the first paper of this series, based on the stellar colors from the Stellar Abundances and Galactic Evolution Survey (SAGES) DR1 and Gaia EDR3. The optimal design of stellar-parameter sensitive $uv$ filters by SAGES has enabled us to determine photometric-metallicity estimates down to $-3.5$, similar to our previous results with the SkyMapper Southern Survey (SMSS), yielding a large sample of over five million metal-poor (MP; [Fe/H]$\le -1.0$) stars and nearly one million very metal-poor (VMP; [Fe/H]$\le -2.0$) stars. The typical precision is around $0.1$ dex for both dwarf and giant stars with [Fe/H]$>-1.0$, and 0.15-0.25/0.3-0.4 dex for dwarf/giant stars with [Fe/H]$<-1.0$. Using the precise parallax measurements and stellar colors from Gaia, effective temperature, luminosity classification, distance and stellar age are further derived for our sample stars. This huge data set in the Northern sky from SAGES, together with similar data in the Southern sky from SMSS, will greatly advance our understanding of the Milky Way, in particular its formation and evolution.

astro-ph.GA↗

arXiv2023

Probing Quasar Viewing Angle with the Variability Structure Function

Given the anisotropic emission from quasar accretion discs, their viewing angle affects estimates of the quasar luminosity, black-hole mass and Eddington ratio. Discs appear overluminous when viewed pole-on and underluminous when viewed at high inclination. In radio-quiet quasars, the viewing angle is usually unknown, although spectroscopic indicators have been proposed. Here, we use a recently discovered universality in the variability structure function (SF) of quasar light curves (LCs), where all quasars show the same SF when clocks run in units of orbital timescale. As an offset from the mean relation can be caused by incorrect orbital timescales and thus incorrect luminosities, we correlate these offsets with suggested inclination indicators. We derive SFs from NASA/ATLAS LCs spanning $\sim 6$ years of observation, using a sample of 183 luminous quasars with measured H$β$ lines as well as 753 quasars with CIV and MgII lines. Starting from the proposed orientation indicators, we expect quasars with narrower H$β$ lines and with more blueshifted CIV lines to be viewed more pole-on and thus appear overluminous. In contrast, our SF analysis finds that presumed pole-on discs appear underluminous, consistently for both line indicators. We discuss possible explanations for the behaviour of quasars with highly blueshifted CIV lines irrespective of inclination angle, including dusty outflows that might render the accretion disc underluminous and flatter disc temperature profiles with longer orbital timescales than in thin-disc models but reach no satisfying conclusion.

astro-ph.GA↗

arXiv2023

Teaching Agents how to Map: Spatial Reasoning for Multi-Object Navigation

In the context of visual navigation, the capacity to map a novel environment is necessary for an agent to exploit its observation history in the considered place and efficiently reach known goals. This ability can be associated with spatial reasoning, where an agent is able to perceive spatial relationships and regularities, and discover object characteristics. Recent work introduces learnable policies parametrized by deep neural networks and trained with Reinforcement Learning (RL). In classical RL setups, the capacity to map and reason spatially is learned end-to-end, from reward alone. In this setting, we introduce supplementary supervision in the form of auxiliary tasks designed to favor the emergence of spatial perception capabilities in agents trained for a goal-reaching downstream objective. We show that learning to estimate metrics quantifying the spatial relationships between an agent at a given location and a goal to reach has a high positive impact in Multi-Object Navigation settings. Our method significantly improves the performance of different baseline agents, that either build an explicit or implicit representation of the environment, even matching the performance of incomparable oracle agents taking ground-truth maps as input. A learning-based agent from the literature trained with the proposed auxiliary losses was the winning entry to the Multi-Object Navigation Challenge, part of the CVPR 2021 Embodied AI Workshop.

cs.CV↗

arXiv2023

Eagle: Large-Scale Learning of Turbulent Fluid Dynamics with Mesh Transformers

Estimating fluid dynamics is classically done through the simulation and integration of numerical models solving the Navier-Stokes equations, which is computationally complex and time-consuming even on high-end hardware. This is a notoriously hard problem to solve, which has recently been addressed with machine learning, in particular graph neural networks (GNN) and variants trained and evaluated on datasets of static objects in static scenes with fixed geometry. We attempt to go beyond existing work in complexity and introduce a new model, method and benchmark. We propose EAGLE, a large-scale dataset of 1.1 million 2D meshes resulting from simulations of unsteady fluid dynamics caused by a moving flow source interacting with nonlinear scene structure, comprised of 600 different scenes of three different types. To perform future forecasting of pressure and velocity on the challenging EAGLE dataset, we introduce a new mesh transformer. It leverages node clustering, graph pooling and global attention to learn long-range dependencies between spatially distant data points without needing a large number of iterations, as existing GNN methods do. We show that our transformer outperforms state-of-the-art performance on, both, existing synthetic and real datasets and on EAGLE. Finally, we highlight that our approach learns to attend to airflow, integrating complex information in a single iteration.

cs.LG↗

arXiv2023

Characterising SMSS J2157--3602, the most luminous known quasar, with accretion disc models

We develop an accretion disc (AD) fitting method, utilising thin and slim disc models and Bayesian inference with the Markov-Chain Monte-Carlo approach, testing it on the most luminous known quasar, SMSS J215728.21-360215.1, at redshift $z=4.692$. With a spectral energy distribution constructed from near-infrared spectra and broadband photometry, the AD models find a black hole mass of $\log(M_{\rm{AD}}/M_{\odot}) = 10.31^{+0.17}_{-0.14}$ with an anisotropy-corrected bolometric luminosity of $\log{(L_{\rm{bol}}/\rm{erg\,s^{-1}})} = 47.87 \pm 0.10$, and derive an Eddington ratio of $0.29^{+0.11}_{-0.10}$ as well as a radiative efficiency of $0.09^{+0.05}_{-0.03}$. Using the near-infrared spectra, we estimate the single-epoch virial black hole mass estimate to be $\log(M_{\rm{SE}}/M_{\odot}) = 10.33 \pm 0.08$, with a monochromatic luminosity at 3000Å of $\log{(L(\rm{3000\textÅ})/\rm{erg\,s^{-1}})} = 47.66 \pm 0.01$. As an independent approach, AD fitting has the potential to complement the single-epoch virial mass method in obtaining stronger constraints on properties of massive quasar black holes across a wide range of redshifts.

astro-ph.GA↗

arXiv2023

AllBRICQS: the All-sky BRIght, Complete Quasar Survey

We describe the first results from the All-sky BRIght, Complete Quasar Survey (AllBRICQS), which aims to discover the last remaining optically bright quasars. We present 156 spectroscopically confirmed quasars (140 newly identified) having |b|>10deg. 152 of the quasars have Gaia DR3 magnitudes brighter than B_P=16.5 or R_P=16 mag, while four are slightly fainter. The quasars span a redshift range of z=0.07-3.93. In particular, we highlight the properties of J0529-4351 at z=3.93, which, if unlensed, is one of the most intrinsically luminous quasars in the Universe. The AllBRICQS sources have been selected by combining data from the Gaia and WISE all-sky satellite missions, and we successfully identify quasars not flagged as candidates by Gaia Data Release 3. We expect the completeness to be approximately 96% within our magnitude and latitude limits, while the preliminary results indicate a selection purity of approximately 96%. The optical spectroscopy used for source classification will also enable detailed quasar characterisation, including black hole mass measurements and identification of foreground absorption systems. The AllBRICQS sources will greatly enhance the number of quasars available for high-signal-to-noise follow-up with present and future facilities.

astro-ph.GA↗

arXiv2023

Universality in the Random Walk Structure Function of Luminous Quasi-Stellar Objects

Rapidly growing black holes are surrounded by accretion disks that make them the brightest objects in the Universe. Their brightness is known to be variable, but the causes of this are not implied by simple disk models and still debated. Due to the small size of accretion disks and their great distance, there are no resolved images addressing the puzzle. In this work, we study the dependence of their variability on luminosity, wavelength and orbital/thermal timescale. We use over 5,000 of the most luminous such objects with light curves of almost nightly cadence from $>5$ years of observations by the NASA/ATLAS project, which provides 2 billion magnitude pairs for a structure function analysis. When time is expressed in units of orbital or thermal time scale in thin-disk models, we find a universal structure function, independent of luminosity and wavelength, supporting the model of magneto-rotational instabilities as a main cause. Over a $>1$~dex range in time, the fractional variability amplitude follows $\log (A/A_0) \simeq 1/2 \times \log (Δt/t_{\rm th})$. Deviations from the universality may hold clues on the structure and orientation of disks.

astro-ph.GA↗

arXiv2022

Spatially Dependent Photometric Activity of M dwarfs in the Solar Cylinder

We study the relationship between Galactic location ($R, Z$) and photometric activity for 3.6 million M dwarf stars within 1 kpc of the Sun. For this purpose, we identify 906 unique flare events as a proxy for magnetic activity from the SkyMapper Southern Survey DR3. We adopt vertical distance $|Z|$ from the Galactic disc as a proxy for age and confirm a strong trend of flaring fraction decreasing with growing stellar age. Among M dwarfs within 50 pc of the Sun, we find a flaring fraction of 1-in-1,500, independent of spectral type from M2 to M7, suggesting that these stars are all in a flare-saturated young evolutionary stage. We find a hint of a kink in the slope of the overall flare fraction near 100 pc from the plane, where a steep decline begins; this slope change is visible for mid-type M dwarfs (M3--M5), suggesting it is not an artefact of mixing spectral type. Together with SDSS H$α$ emission, this trend is additional evidence that the activity fraction of M dwarfs depends on Galactic height and activity lifetime. While there is a hint of flattening of the overall activity fraction above $|Z|\approx$ 500 pc, our data do not constrain this further. Within $\sim$500 pc distance from the Sun, we find no sign of radial disk gradients in flare activity, which may only be revealed by samples covering a larger radial range.

astro-ph.SR↗

arXiv2022

The LSST era of supermassive black holes accretion-disk reverberation mapping

The Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) will detect an unprecedentedly large sample of actively accreting supermassive black holes with typical accretion disk (AD) sizes of a few light days. This brings us to face challenges in the reverberation mapping (RM) measurement of AD sizes in active galactic nuclei (AGNs) using interband continuum delays. We examine the effect of LSST cadence strategies on AD RM using our metric AGNTimeLagMetric. It accounts for redshift, cadence, the magnitude limit, and magnitude corrections for dust extinction. Running our metric on different LSST cadence strategies, we produce an atlas of the performance estimations for LSST photometric RM measurements. We provide an upper limit on the estimated number of quasars for which the AD time lag can be computed within 0 1000 sources in each Deep Drilling field (DDF, 10 sq. deg) in any filter, with the redshift distribution of these sources peaking at z~1. We find the LSST observation strategies with a good cadence (~ 5 days) and a long cumulative season (~9 yr), as proposed for LSST DDF, are favored for the AD size measurement. We create synthetic LSST light curves for the most suitable DDF cadences and determine RM time lags to demonstrate the impact of the best cadences based on the proposed metric.

astro-ph.IM↗