Learning the dynamics of the external world: Brain inspired learning for robotic applications

Humans have exceptional abilities to learn new skills, manipulate tools and objects, and interact with our environment. In order to be successful at these tasks, our brain has become exceptionally well adapted to learning to deal not only with the complex dynamics of our own limbs but also with novel dynamics in the external world. While learning of these dynamics includes learning the complex time-varying forces at the end of limbs through the updating of internal models, it must also include learning the appropriate mechanical impedance in order to stabilize both the limb and any objects contacted in the environment. This article reviews the field of human learning by examining recent experimental evidence about adaptation to novel unstable dynamics and explores how this knowledge about the brain and neuro-muscular system can expand the learning capabilities of robotics and prosthetics. © 2006.

In vivo dynamics of the internal fibrous structure in smooth adhesive pads of insects.

Many insects with smooth adhesive pads can rapidly enlarge their contact area by centripetal pulls on the legs, allowing them to cope with sudden mechanical perturbations such as gusts of wind or raindrops. The short time scale of this reaction excludes any neuromuscular control; it is thus more likely to be caused by mechanical properties of the pad's specialized cuticle. This soft cuticle contains numerous branched fibrils oriented almost perpendicularly to the surface. Assuming a fixed volume of the water-filled cuticle, we hypothesized that pulls could decrease the fibril angle, thereby helping the contact area to expand laterally and longitudinally. Three-dimensional fluorescence microscopy on the cuticle of smooth stick insect pads confirmed that pulls significantly reduced the fibril angle. However, the fibril angle variation appeared insufficient to explain the observed increase in contact area. Direct strain measurements in the contact zone demonstrated that pulls not only expand the cuticle laterally, but also add new contact area at the pad's outer edge.

Generalization in Adaptation to Stable and Unstable Dynamics

Humans skillfully manipulate objects and tools despite the inherent instability. In order to succeed at these tasks, the sensorimotor control system must build an internal representation of both the force and mechanical impedance. As it is not practical to either learn or store motor commands for every possible future action, the sensorimotor control system generalizes a control strategy for a range of movements based on learning performed over a set of movements. Here, we introduce a computational model for this learning and generalization, which specifies how to learn feedforward muscle activity in a function of the state space. Specifically, by incorporating co-activation as a function of error into the feedback command, we are able to derive an algorithm from a gradient descent minimization of motion error and effort, subject to maintaining a stability margin. This algorithm can be used to learn to coordinate any of a variety of motor primitives such as force fields, muscle synergies, physical models or artificial neural networks. This model for human learning and generalization is able to adapt to both stable and unstable dynamics, and provides a controller for generating efficient adaptive motor behavior in robots. Simulation results exhibit predictions consistent with all experiments on learning of novel dynamics requiring adaptation of force and impedance, and enable us to re-examine some of the previous interpretations of experiments on generalization. © 2012 Kadiallah et al.

Wetness loss prediction for a low pressure steam turbine using computational fluid dynamics

Two-phase computational fluid dynamics modelling is used to investigate the magnitude of different contributions to the wet steam losses in a three-stage model low pressure steam turbine. The thermodynamic losses (due to irreversible heat transfer across a finite temperature difference) and the kinematic relaxation losses (due to the frictional drag of the drops) are evaluated directly from the computational fluid dynamics simulation using a concept based on entropy production rates. The braking losses (due to the impact of large drops on the rotor) are investigated by a separate numerical prediction. The simulations show that in the present case, the dominant effect is the thermodynamic loss that accounts for over 90% of the wetness losses and that both the thermodynamic and the kinematic relaxation losses depend on the droplet diameter. The numerical results are brought into context with the well-known Baumann correlation, and a comparison with available measurement data in the literature is given. The ability of the numerical approach to predict the main wetness losses is confirmed, which permits the use of computational fluid dynamics for further studies on wetness loss correlations. © IMechE 2013 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav.

Exploiting passive dynamics for robot throwing task

Throwing is a complex and highly dynamic task. Humans usually exploit passive dynamics of their limbs to optimize their movement and muscle activation. In order to approach human throwing, we developed a double pendulum robotic platform. To introduce passivity into the actuated joints, clutches were included in the drive train. In this paper, we demonstrate the advantage of exploiting passive dynamics in reducing the mechanical work. However, engaging and disengaging the clutches are done in discrete fashions. Therefore, we propose an optimization approach which can deal with such discontinuities. It is shown that properly engaging/disengaging the clutches can reduce the mechanical work of a throwing task. The result is compared to the solution of fully actuated double pendulum, both in simulation and experiment. © 2012 IEEE.

Trajectory control based on discrete full-range dynamics

There has been an increasing interest in the use of mechanical dynamics, (e.g., assive, Elastic, And viscous dynamics) for energy efficient and agile control of robotic systems. Despite the impressive demonstrations of behavioural performance, The mechanical dynamics of this class of robotic systems is still very limited as compared to those of biological systems. For example, Passive dynamic walkers are not capable of generating joint torques to compensate for disturbances from complex environments. In order to tackle such a discrepancy between biological and artificial systems, We present the concept and design of an adaptive clutch mechanism that discretely covers the full-range of dynamics. As a result, The system is capable of a large variety of joint operations, including dynamic switching among passive, actuated and rigid modes. The main innovation of this paper is the framework and algorithm developed for controlling the trajectory of such joint. We present different control strategies that exploit passive dynamics. Simulation results demonstrate a significant improvement in motion control with respect to the speed of motion and energy efficiency. The actuator is implemented in a simple pendulum platform to quantitatively evaluate this novel approach.

Tensile and compressive mechanical behavior of twinned silicon carbide nanowires

Molecular dynamics simulations with the Tersoff potential were used to study the response of twinned SiC nanowires under tensile and compressive strain. The critical strain of the twinned nanowires can be enhanced by twin stacking faults, and their critical strains are larger than those of perfect nanowires with the same diameters. Under axial tensile strain, the bonds of the nanowires are stretched just before failure. The failure behavior is found to depend on the twin segment thickness and the diameter of the nanowires. An atomic chain is observed for thin nanowires with small twin segment thickness under tension strain. Under axial compressive strain, the collapse of twinned SiC nanowires exhibits two different failure modes, depending on the length and diameter of the nanowires, i.e., shell buckling for short nanowires and columnar buckling for longer nanowires.

Mechanical Behavior of Nanometer Ni by Simulating Nanoindentation

An indentation simulation of the crystal Ni is carried out by a molecular dynamics technique (MD) to study the mechanical behavior at nanometer scales. Indenter tips with both sphere shape and conical shape with 60 cone angle are used, and simulation samples with different crystal orientations are adopted. Some defects such as dislocations and point defects are observed. It is found that nucleated defects (dislocations, amorphous atoms) are from the local region near the pin tip or the sample surface. The temperature distribution of the local region is analyzed and it can explain our MD simulation results.

Mechanical Behavior of Nanometer Ni by MD Simulation

The indention simulation of the crystal Ni is carried out by molecular dynamics technique (MD) to study the mechanical behavior at nanometer scales, the indenter tips with sphere shape is used. Some defects such as dislocations, point defects are observed. It is found that defects (dislocations, amorphous) nucleated is from local region near the pin tip or the sample surface. The temperature distribution of local region is analyzed and it can explain our MD simulation result.

Quasi-3D Cytoskeletal Dynamics of Osteocytes under Fluid Flow

Osteocytes respond to dynamic fluid shear loading by activating various biochemical pathways, mediating a dynamic process of bone formation and resorption. Whole-cell deformation and regional deformation of the cytoskeleton may be able to directly regulate this process. Attempts to image cellular deformation by conventional microscopy techniques have been hindered by low temporal or spatial resolution. In this study, we developed a quasi-three-dimensional microscopy technique that enabled us to simultaneously visualize an osteocyte's traditional bottom-view profile and a side-view profile at high temporal resolution. Quantitative analysis of the plasma membrane and either the intracellular actin or microtubule (MT) cytoskeletal networks provided characterization of their deformations over time. Although no volumetric dilatation of the whole cell was observed under flow, both the actin and MT networks experienced primarily tensile strains in all measured strain components. Regional heterogeneity in the strain field of normal strains was observed in the actin networks, especially in the leading edge to flow, but not in the MT networks. In contrast, side-view shear strains exhibited similar subcellular distribution patterns in both networks. Disruption of MT networks caused actin normal strains to decrease, whereas actin disruption had little effect on the MT network strains, highlighting the networks' mechanical interactions in osteocytes.