Alberto Badias

Universidad Politécnica de Madrid | UPM

El objetivo de este trabajo es desarrollar herramientas de simulación en tiempo real en el contexto de la realidad aumentada. Para ello, se hace uso de los algoritmos de inteligencia artificial desarrollados en trabajos previos que aseguran la consistencia física de los resultados.

El objetivo de este trabajo es la utilización de redes neuronales para obtenener soluciones de alta resolución de un sistema dinámico cuando únicamente se dispone de medidas parciales. Para ello se utilizan sesgos inductivos que aplican los principios de la termodinámica, así como la aplicación de técnicas de superresolución.

We develop inductive biases for the machine learning of complex physical systems based on the port-Hamiltonian formalism. To satisfy by construction the principles of thermodynamics in the learned physics (conservation of energy, non-negative entropy production), we modify accordingly the port-Hamiltonian formalism so as to achieve a port-metriplec...

Learning and reasoning about physical phenomena is still a challenge in robotics development, and computational sciences play a capital role in the search for accurate methods able to provide explanations for past events and rigorous forecasts of future situations. We propose a thermodynamics-informed active learning strategy for fluid perception a...

We present a new framework to measure the intrinsic properties of (deep) neural networks. While we focus on convolutional networks, our framework can be extrapolated to any network architecture. In particular, we evaluate two network properties, namely, capacity, which is related to expressivity, and compression, which is related to learnability. B...

We develop inductive biases for the machine learning of complex physical systems based on the port-Hamiltonian formalism. To satisfy by construction the principles of thermodynamics in the learned physics (conservation of energy, non-negative entropy production), we modify accordingly the port-Hamiltonian formalism so as to achieve a port-metriplec...

The imminent impact of immersive technologies in society urges for active research in real-time and interactive physics simulation for virtual worlds to be realistic. In this context, realistic means to be compliant to the laws of physics. In this paper we present a method for computing the dynamic response of (possibly non-linear and dissipative)...

Physics perception very often faces the problem that only limited data or partial measurements on the scene are available. In this work, we propose a strategy to learn the full state of sloshing liquids from measurements of the free surface. Our approach is based on recurrent neural networks (RNN) that project the limited information available to a...

Learning and reasoning about physical phenomena is still a challenge in robotics development, and computational sciences play a capital role in the search for accurate methods able to provide explanations for past events and rigorous forecasts of future situations. We propose a physics-informed reinforcement learning strategy for fluid perception a...

In this paper we present a deep learning method to predict the time evolution of dissipative dynamical systems. We propose using both geometric and thermodynamic inductive biases to improve accuracy and generalization of the resulting integration scheme. The first is achieved with Graph Neural Networks, which induces a non-Euclidean geometrical pri...

In this paper we present a deep learning method to predict the temporal evolution of dissipative dynamic systems. We propose using both geometric and thermodynamic inductive biases to improve accuracy and generalization of the resulting integration scheme. The first is achieved with Graph Neural Networks, which induces a non-Euclidean geometrical p...

El objetivo de este trabajo es utilizar inteligencia artificial para aprender las leyes físicas que gobiernan un sistema arbitrario y predecir así su evolución en el tiempo. Para ello se utilizan varios sesgos inductivos, que aseguran la correcta estructura matemática del problema y mejoran su capacidad de generalización.

We propose a new methodology to estimate the 3D displacement field of deformable objects from video sequences using standard monocular cameras. We solve in real time the complete (possibly visco-)hyperelasticity problem to properly describe the strain and stress fields that are consistent with the displacements captured by the images, constrained b...

We present a new framework to measure the intrinsic properties of (deep) neural networks. While we focus on convolutional networks, our framework can be extrapolated to any network architecture. In particular, we evaluate two network properties, namely, capacity (related to expressivity) and compression, both of which depend only on the network str...

Physics perception very often faces the problem that only limited data or partial measurements on the scene are available. In this work, we propose a strategy to learn the full state of sloshing liquids from measurements of the free surface. Our approach is based on recurrent neural networks (RNN) that project the limited information available to a...

We present an algorithm to learn the relevant latent variables of a large-scale discretized physical system and predict its time evolution using thermodynamically-consistent deep neural networks. Our method relies on sparse autoencoders, which reduce the dimensionality of the full order model to a set of sparse latent variables with no prior knowle...

We have seen in Part I of this paper that model order reduction allows the involvement of physics-based models in design, as in the past, but now also in online decision-making, without requiring unreasonable computing resources. On the other hand, machine learning techniques were not ready to cope with the processing speed and the lack of data. It...

We are interested in evaluating the state of drivers to determine whether they are attentive to the road or not by using motion sensor data collected from car driving experiments. That is, our goal is to design a predictive model that can estimate the state of drivers given the data collected from motion sensors. For that purpose, we leverage recen...

En este trabajo se pretende aplicar inteligencia artificial para predecir la evolución temporal de un sistema físico arbitrario sólo mediante el uso de datos, es decir, sin conocer las ecuaciones que lo rigen. Este método se ha validadado con dos sistemas dinámicos disipativos: uno continuo y otro discreto.

We develop a method to learn physical systems from data that employs feedforward neural networks and whose predictions comply with the first and second principles of thermodynamics. The method employs a minimum amount of data by enforcing the metriplectic structure of dissipative Hamiltonian systems in the form of the so-called General Equation for...

We propose a new methodology to estimate the 3D displacement field of deformable objects from video sequences using standard monocular cameras. We solve in real time the complete (possibly visco-)hyperelasticity problem to properly describe the strain and stress fields that are consistent with the displacements captured by the images, constrained b...

Digital twins can be defined as digital representations of physical entities that employ real‐time data to enable understanding of the operating conditions of these entities. Here we present a particular type of digital twin that involves a combination of computer vision, scientific machine learning and augmented reality. This novel digital twin is...

We present an algorithm to learn the relevant latent variables of a large-scale discretized physical system and predict its time evolution using thermodynamically-consistent deep neural networks. Our method relies on sparse autoencoders, which reduce the dimensionality of the full order model to a set of sparse latent variables with no prior knowle...

We present a real‐time method for computing the mechanical interaction between real and virtual objects in an augmented reality environment. Using model order reduction methods we are able to estimate the physical behaviour of deformable objects in real time, with the precision of a high‐fidelity solver but working at the speed of a video sequence....

We develop a method to learn physical systems from data that employs feedforward neural networks and whose predictions comply with the first and second principles of thermodynamics. The method employs a minimum amount of data by enforcing the metriplectic structure of dissipative Hamiltonian systems in the form of the so-called General Equation for...

While modern CFD tools are able to provide the user with reliable and accurate simulations, there is a strong need for interactive design and analysis tools. State of the art CFD software employs massive resources in terms of CPU time, user interaction and also GPU time for rendering and analysis. In this work we develop an innovative tool able to...

We present a new way of adding augmented information based on the computation of the physical equations that truly govern the behavior of objects. In computer graphics, it is common to use big simplifications to be able to solve this type of equations in real time, obtaining in many occasions behaviors that differ remarkably from reality. However,...

In this work we explore the possibilities of reduced order modeling for augmented reality applications. We consider parametric reduced order models based upon separate (affine) parametric dependence so as to speedup the associated data assimilation problems, which involve in a natural manner the minimization of a distance functional. The employ of...

Augmented reality is one of the fields with greatest interest in technological research. Real time requirements force to use physics engines to approximate the behaviour of the objects. We propose the computation of the proper equations that govern the physics of deformable objects, and their interaction with users in real time, using dimensionalit...

Local model order reduction methods provide better results than global ones to problems with intricate manifold solution structure. A posteriori methods (e.g. Proper Orthogonal Decomposition) have been many times applied locally, but a priori methods (e.g. Proper Generalized Decomposition) have the difficulty of determining the manifold structure o...

Data assimilation is the process by which experimental measurements are incorporated into the modeling process of a given system. We focus here on the framework of non-linear solid mechanics. Applications of the developed methodology include real-time monitoring and control of structures or mixed/augmented reality, to name a few. In these circumsta...

Recent research on visual prosthesis demonstrates the possibility of providing visual perception to people with certain blindness. Bypassing the damaged part of the visual path, electrical stimulation provokes spot percepts known as phosphenes. Due to physiological and technological limitations the information received by patients has very low reso...

One of the main difficulties that a reduced‐order method could face is the poor separability of the solution. This problem is common to both a posteriori model order reduction (proper orthogonal decomposition, reduced basis) and a priori [proper generalized decomposition (PGD)] model order reduction. Early approaches to solve it include the constru...