Search

Robots in human-centered environments require accurate scene understanding to perform high-level tasks effectively. This understanding can be achieved through instance-aware semantic mapping, which involves reconstructing elements at the level of individual instances. Neural networks, the de facto solution for scene understanding, still face limitations such as overconfident incorrect predictions with out-of-distribution objects or generating inaccurate masks.Placing excessive reliance on these predictions makes the reconstruction susceptible to errors, reducing the robustness of the resulting maps and hampering robot operation. In this work, we propose Voxeland, a probabilistic framework for incrementally building instance-aware semantic maps. Inspired by the Theory of Evidence, Voxeland treats neural network predictions as subjective opinions regarding map instances at both geometric and semantic levels. These opinions are aggregated over time to form evidences, which are formalized through a probabilistic model. This enables us to quantify uncertainty in the reconstruction process, facilitating the identification of map areas requiring improvement (e.g. reobservation or reclassification). As one strategy to exploit this, we incorporate a Large Vision-Language Model (LVLM) to perform semantic level disambiguation for instances with high uncertainty. Results from the standard benchmarking on the publicly available SceneNN dataset demonstrate that Voxeland outperforms state-of-the-art methods, highlighting the benefits of incorporating and leveraging both instance- and semantic-level uncertainties to enhance reconstruction robustness. This is further validated through qualitative experiments conducted on the real-world ScanNet dataset.

The behavior of velocity fluctuations near a wall has long fascinated the turbulence community, because the prevalent theoretical framework of an attached-eddy hierarchy appears to predict infinite intensities as the Reynolds number tends to infinity. Although an unbounded infinite limit is not a problem in itself, it raises the possibility of unfamiliar phenomena when the Reynolds number is large, and has motivated attempts to avoid it. We review the subject and point to possible pitfalls stemming from uncritical extrapolation from low Reynolds numbers, or from an over-simplification of the multiscale nature of turbulence. It is shown that large attached eddies dominate the high-Reynolds-number regime of the near-wall layer, and that they behave differently from smaller-scale ones. In that limit, the near-wall layer is controlled by the outer flow, the large-scale fluctuations reduce to turbulent layers modulated by a variable friction velocity, and the kinetic-energy peak is substituted by a deeper structure with a secondary outer maximum. The friction velocity is then not necessarily the best velocity scale. While the near-wall energy peak probably becomes unbounded in wall units, it tends to zero when expressed in terms of the outer velocity.

The causal relevance of local flow conditions in wall-bounded turbulence is analysed using ensembles of interventional experiments in which the effect of perturbing the flow within a small cell is monitored at some future time. When this is done using the relative amplification of the perturbation energy, causality depends on the flow conditions within the cell before it is perturbed, and can be used as a probe of the flow dynamics. The key scaling parameter is the ambient shear, which is also the dominant diagnostic variable for wall-attached perturbations. Away from the wall, the relevant variables are the streamwise and wall-normal velocities. Causally significant cells are associated with sweeps that carry the perturbation towards the stronger shear near the wall, whereas irrelevant ones are associated with ejections that carry it towards the weaker shear in the outer layers. Causally significant and irrelevant cells are themselves organised into structures that share many characteristics with classical sweeps and ejections, such as forming spanwise pairs whose dimensions and geometry are similar to those of classical quadrants. At the wall, this is consistent with causally significant configurations in which a high-speed streak overtakes a low-speed one, and causally irrelevant ones in which the two streaks pull apart from each other. It is argued that this is probably associated with streak meandering.

High-dimensional dynamical systems projected onto a reduced-order model cease to be deterministic and are best described by probability distributions in state space. Their equations of motion map onto an evolution operator with a deterministic component describing the projected dynamics, and a stochastic one from the neglected dimensions. It is shown that, for projections in which the deterministic component is dominant, `physics-free' stochastic Markovian models can be constructed that mimic many of the statistics of the real flow, even for fairly crude operator approximations. Deterministic models converge to steady states. This is related to general properties of Markov chains and illustrated with data-driven models for a moderate-Reynolds number turbulent channel.

We present a novel 3D odometry method that recovers the full motion of a vehicle only from a Doppler-capable range sensor. It leverages the radial velocities measured from the scene, estimating the sensor's velocity from a single scan. The vehicle's 3D motion, defined by its linear and angular velocities, is calculated taking into consideration its kinematic model which provides a constraint between the velocity measured at the sensor frame and the vehicle frame. Experiments carried out prove the viability of our single-sensor method compared to mounting an additional IMU. Our method provides the translation of the sensor, which cannot be reliably determined from an IMU, as well as its rotation. Its short-term accuracy and fast operation (~5ms) make it a proper candidate to supply the initialization to more complex localization algorithms or mapping pipelines. Not only does it reduce the error of the mapper, but it does so at a comparable level of accuracy as an IMU would. All without the need to mount and calibrate an extra sensor on the vehicle., Comment: This work has been submitted to the IEEE for possible publication

Gas source localization (GSL) with an autonomous robot is a problem with many prospective applications, from finding pipe leaks to emergency-response scenarios. In this work, we present a new method to perform GSL in realistic indoor environments, featuring obstacles and turbulent flow. Given the highly complex relationship between the source position and the measurements available to the robot (the single-point gas concentration, and the wind vector) we propose an observation model that derives from contrasting the online, real-time simulation of the gas dispersion from any candidate source localization against a gas concentration map built from sensor readings. To account for a convenient and grounded integration of both into a probabilistic estimation framework, we introduce the concept of probabilistic gas-hit maps, which provide a higher level of abstraction to model the time-dependent nature of gas dispersion. Results from both simulated and real experiments show the capabilities of our current proposal to deal with source localization in complex indoor environments.

Flow patterns of causal significance to three-dimensional isotropic turbulence are identified through the recently introduced algorithm of Jim\'enez (2018). Localised perturbations are introduced at arbitrary regions of a triple-periodic decaying flow at ${Re}_\lambda \approx 190$, and their evolution is used as a marker of the significance of said regions to the flow. Their dimensions are found to be an important parameter, with sizes of the order of the integral scale being controlled by the kinetic energy content, and sizes within the dissipative range, by the enstrophy and dissipation. The three quantities are found to be important at intermediate (inertial) scales. Prominent differences emerge between regions of high and low significance. The former typically contain strong gradients and/or kinetic energy and the latter are weak. An analysis of the structure of significant and insignificant flow patterns reveals that strain is more efficient than vorticity at propagating the contents of the perturbation to other regions of the flow. Moreover, the flow patterns of significant regions are found to be more complex, typically containing vortex clusters, while simpler vortex sheets are found in insignificant regions. The present results suggest that strategies aiming to manipulate the flow should focus on strain-dominated vortex clusters, avoiding enstrophy-dominated vortex sheets. This is confirmed through an assimilation experiment, in which greater synchronisation between two simulations is achieved when simulations share significant regions rather than insignificant ones. These conclusions have implications for both the control of turbulent flows and the making of predictions based on limited or noisy measurements., Comment: 39 pages, 22 figures

The Perron-Frobenius operator (PFO) is adapted from dynamical-system theory to the study of turbulent channel flow. It is shown that, as long as the analysis is restricted to the system attractor, the PFO can be used to differentiate causality and coherence from simple correlation without performing interventional experiments, and that the key difficulty remains collecting enough data to populate the operator matrix. This is alleviated by limiting the analysis to two-dimensional projections of the phase space, and developing a series of indicators to choose the best parameter pairs from a large number of possibilities. The techniques thus developed are applied to the study of bursting in the inertial layer of the channel, with emphasis on the process by which bursts are reinitiated after they have decayed. Conditional averaging over phase-space trajectories suggested by the PFO shows, somewhat counter-intuitively, that a key ingredient for the burst recovery is the development of a low-shear region near the wall, overlaid by a lifted shear layer. This is confirmed by a computational experiment in which the control of the mean velocity profile by the turbulence fluctuations is artificially relaxed. The behaviour of the mean velocity profile is thus modified, but the association of low wall shear with the initiation of the bursts is maintained.

The interactions between the near-wall streaks and the large-scale motions (LSMs) of the outer region of wall-bounded turbulent flows are investigated. The co-supporting hypothesis of Toh & Itano (2005) is checked in full-scale channels at low to moderate Reynolds numbers, from two points of view. To study the top-down influence of the outer structures on the spanwise motion of the near-wall streaks, a method inspired by particle-image velocimetry is used to track the spanwise position of the streaks. Their spanwise advection velocity is found to be affected by the hierarchy of large-scale circulations in the logarithmic layer, but their spanwise streak density is only weakly related to the LSMs. The evidence suggests that a top-down influence exists and drives the drift of the streaks in the spanwise direction, as suggested by Toh & Itano (2005), but that the hypothesized streak accumulation rarely occurs. Numerical experiments at Re_tau about 535 are then performed to clarify the role of the near-wall streaks in the generation and preservation of the outer LSMs. The results show that the merger of the near-wall streaks is only weakly correlated with the generation of the LSMs, and that removing the near-wall roots of the LSMs does not affect the evolution of their outer region. It is concluded that the bottom-up influence from the near-wall streaks is not essential for the LSM generation and preservation, also weakening the evidence for the co-supporting hypothesis.

The effect of damping the longest streaks in wall-bounded turbulence is explored using numerical experiments. It is found that long streaks are not required for the self-sustenance of the bursting process, which is relatively little affected by their absence. In particular, there are turbulence states in which the fluctuations of the streamwise velocity have approximately the same length as the bursts, and are thus presumably associated with the bursts themselves, while the burst structure is essentially indistinguishable from flows in which longer velocity fluctuations are present. This suggests that the long streaks found in unmodified flows may be byproducts, rather than active parts of the energy generation cycle.

Opposition flow control is a robust strategy that has been proved effective in turbulent wall-bounded flows. Its conventional setup consists of measuring wall-normal velocity in the buffer layer and opposing it at the wall. This work explores the possibility of implementing this strategy with a detection plane in the logarithmic layer, where control could be feasible experimentally. We apply control on a channel flow at $Re_\tau = 932$, only on the eddies with relatively large wavelengths ($\lambda / h > 0.1$). Similarly to the buffer layer opposition control, our control strategy results in a virtual-wall effect for the wall-normal velocity, creating a minimum in its intensity. However, it also induces a large response in the streamwise velocity and Reynolds stresses near the wall, with a substantial drag increase. When the phase of the control lags with respect to the detection plane, spanwise-homogeneous rollers are observed near the channel wall. We show that they are a result of a linear instability. In contrast, when the control leads with respect to the detection plane, this instability is inactive and oblique waves are observed. Their wall-normal profiles can be predicted linearly as a response of the turbulent channel flow to a forcing with the advection velocity of the detection plane. The linearity, governing the flow, opens a possibility to affect large scales of the flow in a controlled manner, when enhanced turbulence intensity or mixing is desired.