Rapid Concept Learning for Mobile Robots

Concept learning in robotics is an extremely challenging problem: sensory data is often high-dimensional, and noisy due to specularities and other irregularities. In this paper, we investigate two general strategies to speed up learning, based on spatial decomposition of the sensory representation, and simultaneous learning of multiple classes using a shared structure. We study two concept learning scenarios: a hallway navigation problem, where the robot has to induce features such as opening or wall. The second task is recycling, where the robot has to learn to recognize objects, such as a trash can. We use a common underlying function approximator in both studies in the form of a feedforward neural network, with several hundred input units and multiple output units. Despite the high degree of freedom afforded by such an approximator, we show the two strategies provide sufficient bias to achieve rapid learning. We provide detailed experimental studies on an actual mobile robot called PAVLOV to illustrate the effectiveness of this approach.     

On the numerical solution of generalized convection heat transfer problems via the method of proper closure equations – part II: Application to test problems

ABSTRACT

In part I of this paper, a new physical-based computational approach for the solution of convection heat transfer problems on co-located non-orthogonal grids in the context of an element-based finite volume method was discussed. The test problems are presented here, in part II of the paper. These problems include five steady two-dimensional convection heat transfer problems. In all test cases, the convergence history, the required under-relaxations for the iterative solution of the linearized equations, and the order of accuracy of the method are discussed and the streamlines as well as isotherms are presented. The computational results show that the proposed method is second order accurate and might occasionally need mild under-relaxation in relatively complex problems. Excellent match between the computational results and the corresponding reliable published results is observed.     

Aspect ratio effects of an adiabatic rectangular obstacle on natural convection and entropy generation of a nanofluid in an enclosure

In the present study, aspect ratio (AR) effects of a centered adiabatic rectangular obstacle numerically investigated on natural convection and entropy generation in a differentially heated enclosure filled with either water or nanofluid (Cu-water). The governing equations are solved numerically with finite volume method using the SIMPLER algorithm. The study has been done for Rayleigh numbers between 10³ and 10⁶, the aspect ratio of 1/3, 1/2, 1, 2 and 3 and for base fluid as well as nanofluid. It is found that, using the nanofluid leads to increase the flow strength, average Nusselt number and entropy generation and decrease the Bejan number especially at high Rayleigh numbers. At low Rayleigh numbers entropy generation is very low. By increasing Rayleigh number, entropy generation and Bejan number increases. It is observed that the viscose entropy generation is more considerable than the thermal entropy generation and has dominant role in total entropy generation. The maximum entropy generation occurs at AR = 1/3 and 3 and the minimum entropy generation occurs at AR = 1 and 1/2. It is observed that the effect of AR on Nusselt number, entropy generation and Bejan number depends on Rayleigh number.     

Rapid Concept Learning for Mobile Robots

Concept learning in robotics is an extremely challenging problem: sensory data is often high dimensional, and noisy due to specularities and other irregularities. In this paper, we investigate two general strategies to speed up learning, based on spatial decomposition of the sensory representation, and simultaneous learning of multiple classes using a shared structure. We study two concept learning scenarios: a hallway navigation problem, where the robot has to induce features such as opening or wall. The second task is recycling, where the robot has to learn to recognize objects, such as a trash can. We use a common underlying function approximator in both studies in the form of a feedforward neural network, with several hundred input units and multiple output units. Despite the high degree of freedom afforded by such an approximator, we show the two strategies provide sufficient bias to achieve rapid learning. We provide detailed experimental studies on an actual mobile robot called PAVLOV to illustrate the effectiveness of this approach.     

Numerical study of natural convection and entropy generation of Cu-water nanofluid around an obstacle in a cavity

In this work, natural convection and entropy generation in a square cavity with an obstacle filled with Cu-water nanofluid is numerically studied. Horizontal walls of the cavity are adiabatic and vertical walls are maintained at a different constant temperature. The study has been done for the Rayleigh numbers between 10³ and 10⁶, the obstacle dimensions (W/L) of 0.1?C0.5 and for base fluid as well as nanofluid. It is found that, using the nanofluid overall leads to increase the flow strength, Nusselt number and entropy generation and decrease the Bejan number especially at high Rayleigh numbers. It is observed that by increasing the obstacle dimensions, the entropy generation increases and the Bejan number decreases, but the effect of the obstacle dimensions on Nusselt number depends on Rayleigh number. For the present thermal system, the increasing Nusselt number compared to increasing entropy generation due to increase obstacle dimensions is significant at low Rayleigh numbers.     

Antibacterial activities of novel nanocomposite biofilms of chitosan/phosphoramide/Ag NPs

Novel nanocomposite films of chitosan/phosphoramide/Ag NPs were prepared containing 1–5% of silver nanoparticles. The Ag NPs were synthesized according to the citrate reduction method. The XRD and SEM analysis of Ag NPs, chitosan (CS), phosphoramide (Ph), CS/Ph, CS/Ag NPs films and the nanocomposite films 1–5 containing CS/Ph/1–5% Ag NPs were investigated. The in vitro antibacterial activities were evaluated against four bacteria including two Gram‐positive Staphylococcus aureus (S. aureus), Bacillus cereus (B. cereus) and two Gram‐negative Escherchia coli (E. coli), Pseudomonas aeruginosa (P. aeruginosa) bacteria. Results revealed greater antibacterial effects of the films against Gram‐positive bacteria. Also, nanocomposite films containing higher percent of Ag NPs showed more antibacterial activities. POLYM. COMPOS. 36:454–466, 2015. © 2014 Society of Plastics Engineers     

On the numerical solution of generalized convection heat transfer problems via the method of proper closure equations – part I: Description of the method

ABSTRACT

The purpose of this paper is to introduce a new physical-based computational approach for the solution of convection heat transfer problems on co-located non-orthogonal grids in the context of an element-based finite volume method. The approach has already been presented in the context of two-dimensional incompressible flow problems without heat transfer. It has been shown that the pressure–velocity coupling on co-located grids can be correctly modeled via the so-called method of proper closure equations (MPCE). Here, MPCE is extended to the numerical simulation of natural, forced, and mixed convection heat transfer problems. It is shown that the couplings between pressure, velocity, and temperature can be conveniently handled on co-located grids by resorting again to the modified forms of the governing equations, i.e., the proper closure equations. The set of discrete equations is solved in a fully coupled manner in this study. Here, in part I of the paper, only the basic methodology is described; in part II, the results of application of the method to some test problems are presented.     

A coupled element-based finite-volume method for the solution of incompressible Navier-Stokes equations

ABSTRACT

This article presents a new element-based finite-volume discretization approach for the solution of incompressible flow problems on co-located grids. The proposed method, called the method of proper closure equations (MPCE), employs a proper set of physically relevant equations to constrain the velocity and pressure at integration points. These equations provide a proper coupling between the nodal values of pressure and velocity. The final algebraic equations are not segregated in this study and are solved in a fully coupled manner. To show the applicability and performance of the method, it is tested on several steady two-dimensional laminar-flow benchmark cases. The results indicate that the method simulates the fluid flow in complex geometries and on nonorthogonal computational grids accurately. Also, it is shown that the method is robust in the sense that it does not require severe underrelaxation even at relatively high Reynolds numbers. In each test case, the required underrelaxation parameter, the number of iterations, and the corresponding CPU time are reported.     

Numerical investigation of NOx reduction in a sudden-expansion combustor with inclined turbulent air jet

Axisymmetric sudden-expansion geometry of a co-flowing methane-air diffusion flame is considered to investigate the effect of air inlet conditions on NO_x formation, flow field and temperature distribution using the k-? turbulence and β-PDF combustion model. The predicted results are in acceptable agreement with the published experimental and numerical data. The obtained results show that increasing air turbulence intensity results in considerable decrease in NO formation. Increasing the inlet angle of the air causes the NO formation to decrease due to raising vorticity strength. As a new index, the mass-averaged integral of vorticity magnitude is introduced to investigate the effect of altering inlet angle of the air on the flow field.