Computational motor control in humans and robots

Computational models can provide useful guidance in the design of behavioral and neurophysiological experiments and in the interpretation of complex, high dimensional biological data. Because many problems faced by the primate brain in the control of movement have parallels in robotic motor control, models and algorithms from robotics research provide useful inspiration, baseline performance, and sometimes direct analogs for neuroscience.     

Grounding word learning in space

Humans and objects, and thus social interactions about objects, exist within space. Words direct listeners' attention to specific regions of space. Thus, a strong correspondence exists between where one looks, one's bodily orientation, and what one sees. This leads to further correspondence with what one remembers. Here, we present data suggesting that children use associations between space and objects and space and words to link words and objects--space binds labels to their referents. We tested this claim in four experiments, showing that the spatial consistency of where objects are presented affects children's word learning. Next, we demonstrate that a process model that grounds word learning in the known neural dynamics of spatial attention, spatial memory, and associative learning can capture the suite of results reported here. This model also predicts that space is special, a prediction supported in a fifth experiment that shows children do not use color as a cue to bind words and objects. In a final experiment, we ask whether spatial consistency affects word learning in naturalistic word learning contexts. Children of parents who spontaneously keep objects in a consistent spatial location during naming interactions learn words more effectively. Together, the model and data show that space is a powerful tool that can effectively ground word learning in social contexts.     

Dynamical systems for predictive control of autonomous robots

Summary Regularities in the environment are accessible to an autonomous agents as reproducible relations between actions and perceptions and can be exploited by unsupervised learning. Our approach is based on the possibility to perform and to verify predictions about perceivable consequences of actions. It is implemented as a three-layer neural network that combines predictive perception, internal-state transitions and action selection into a loop which closes via the environment. In addition to minimizing prediction errors, the goal of network adaptation comprises also an optimization of the minimization rate such that new behaviors are favored over already learned ones, which would result in a vanishing improvement of predictability. Previously learned behaviors are reactivated or continued if triggering stimuli are available and an externally or otherwise given reward overcompensates the decay of the learning rate. In the model, behavior learning and learning behavior are brought about by the same mechanism, namely the drive to continuously experience learning success. Behavior learning comprises representation and storage of learned behaviors and finally their inhibition such that a further exploration of the environment is possible. Learning behavior, in contrast, detects the frontiers of the manifold of learned behaviors and provides estimates of the learnability of behaviors leading outwards the field of expertise. The network module has been implemented in a Khepera miniature robot. We also consider hierarchical architectures consisting of several modules in one agent as well as groups of several agents, which are controlled by such networks.     

Adaptive control of mobile robots using a neural network

A Neural Network - based control approach for mobile robot is proposed. The weight adaptation is made on-line, without previous learning. Several possible situations in robot navigation are considered, including uncertainties in the model and presence of disturbance. Weight adaptation laws are presented as well as simulation results.     

SVM-based algorithm for recognition of QRS complexes in electrocardiogram

Among all electrocardiogram (ECG) components, the QRS complex is the most significant feature. This paper presents a new algorithm for recognition of QRS complexes in the electrocardiogram (ECG) based on support vector machine (SVM). Digital filtering techniques are used to remove power line interference and baseline wander in the ECG signal. SVM is used as a classifier to delineate QRS and non-QRS regions. Algorithm performance was evaluated against the standard CSE ECG database. The results indicated that the algorithm achieved 99.3% of the detection rate. The percentage of false positive and false negative was 12.4 and 0.7% respectively. It could function reliably even under the condition of poor signal quality of the ECG signal.     

Coordinates transformation and learning control for visually-guided voluntary movement with iteration: A Newton-like method in a function space

In order to control visually-guided voluntary movements, the central nervous system (CNS) must solve the following three computational problems at different levels: (1) determination of a desired trajectory in the visual coordinates, (2) transformation of the coordinates of the desired trajectory to the body coordinates and (3) generation of motor command. In this paper, the second and the third problems are treated at computational, representational and hardware levels of Marr. We first study the problems at the computational level, and then propose an iterative learning scheme as a possible algorithm. This is a trial and error type learning such as repetitive training of golf swing. The amount of motor command needed to coordinate activities of many muscles is not determined at once, but in a step-wise, trial and error fashion in the course of a set of repetitions. Actually, the motor command in the (n+1)-th iteration is a sum of the motor command in then-th iteration plus two modification terms which are, respectively, proportional to acceleration and speed errors between the desired trajectory and the realized trajectory in then-th iteration. We mathematically formulate this iterative learning control as a Newton-like method in functional spaces and prove its convergence under appropriate mathematical conditions with use of dynamical system theory and functional analysis. Computer simulations of this iterative learning control of a robotic manipulator in the body or visual coordinates are shown. Finally, we propose that areas 2, 5, and 7 of the sensory association cortex are possible sites of this learning control. Further we propose neural network model which acquires transformation matrices from acceleration or velocity to motor command, which are used in these schemes.     

Prediction precedes control in motor learning

Skilled motor behavior relies on the brain learning both to control the body and predict the consequences of this control. Prediction turns motor commands into expected sensory consequences, whereas control turns desired consequences into motor commands. To capture this symmetry, the neural processes underlying prediction and control are termed the forward and inverse internal models, respectively. Here, we investigate how these two fundamental processes are related during motor learning. We used an object manipulation task in which subjects learned to move a hand-held object with novel dynamic properties along a prescribed path. We independently and simultaneously measured subjects' ability to control their actions and to predict their consequences. We found different time courses for predictor and controller learning, with prediction being learned far more rapidly than control. In early stages of manipulating the object, subjects could predict the consequences of their actions, as measured by the grip force they used to grasp the object, but could not generate appropriate actions for control, as measured by their hand trajectory. As predicted by several recent theoretical models of sensorimotor control, our results indicate that people can learn to predict the consequences of their actions before they can learn to control their actions.     

Quality control of protein folding in extracellular space

The pathologies of many serious human diseases are thought to develop from the effects of intra- or extracellular aggregates of non-native proteins. Inside cells, chaperone and protease systems regulate protein folding; however, little is known about any corresponding mechanisms that operate extracellularly. The identification of these mechanisms is important for the development of new disease therapies. This review briefly discusses the consequences of protein misfolding, the intracellular mechanisms that control folding and the potential corresponding extracellular control processes. Finally, a new speculative model is described, which proposes that newly discovered extracellular chaperones bind to exposed regions of hydrophobicity on non-native, extracellular proteins to target them for receptor-mediated endocytosis and intracellular, lysosomal degradation.     

Interactive language learning by robots: the transition from babbling to word forms

The advent of humanoid robots has enabled a new approach to investigating the acquisition of language, and we report on the development of robots able to acquire rudimentary linguistic skills. Our work focuses on early stages analogous to some characteristics of a human child of about 6 to 14 months, the transition from babbling to first word forms. We investigate one mechanism among many that may contribute to this process, a key factor being the sensitivity of learners to the statistical distribution of linguistic elements. As well as being necessary for learning word meanings, the acquisition of anchor word forms facilitates the segmentation of an acoustic stream through other mechanisms. In our experiments some salient one-syllable word forms are learnt by a humanoid robot in real-time interactions with naive participants. Words emerge from random syllabic babble through a learning process based on a dialogue between the robot and the human participant, whose speech is perceived by the robot as a stream of phonemes. Numerous ways of representing the speech as syllabic segments are possible. Furthermore, the pronunciation of many words in spontaneous speech is variable. However, in line with research elsewhere, we observe that salient content words are more likely than function words to have consistent canonical representations; thus their relative frequency increases, as does their influence on the learner. Variable pronunciation may contribute to early word form acquisition. The importance of contingent interaction in real-time between teacher and learner is reflected by a reinforcement process, with variable success. The examination of individual cases may be more informative than group results. Nevertheless, word forms are usually produced by the robot after a few minutes of dialogue, employing a simple, real-time, frequency dependent mechanism. This work shows the potential of human-robot interaction systems in studies of the dynamics of early language acquisition.     

Cellular dimensions and cell dynamics, or the difficulty over capturing time and space in the era of electron microscopy

The introduction of electron microscopy profoundly altered biomedical research, providing a tool for a more detailed but at the same time a spatially and temporally more restricted visual analysis. Examining the case study of Golgi apparatus research in the 1950s and 1960s, it will be shown how microscopists handled these challenges, and how these confrontations modified the general concept of cellular organization. This will also shed light on the artifact debate and on the question of scientific realism in the field of microscopy.     

A knowledge-based system for the strategic control level of robots in flexible manufacturing cells

The strategic control level synthesis for robots is related to a hierarchical robot control problem. The main control problem at the strategic control level is to select the model and algorithm to be used by the lower control level to execute the given robot task. Usually there are several lower control level models and algorithms that can be used by the robot control system for every robot task. Strategic control level synthesis depends on the particular robot system application. In a typical application, when the robot system is used in a flexible manufacturing system for manipulating various part types, the robot tasks executed by the robot system depend on the manufacturing processes in the system. If the robot system is applied in another flexible manufacturing system, dedicated to other manufacturing processes, another set of robot tasks might be needed to perform the necessary operations. Therefore, the quantity and the kind of knowledge required in the system for the strategic control level differ from one application to another. Such a fact creates the appropriate conditions for employing some artificial intelligence techniques. This article describes a knowledge-based system approach to the strategic control level synthesis problem.     

Aerial locomotion in flies and robots: kinematic control and aerodynamics of oscillating wings

Flight in flies results from a feedback cascade in which the animal converts mechanical power produced by the flight musculature into aerodynamic forces. A major goal of flight research is to understand the functional significance of the various components in this cascade ranging from the generation of the neural code, the control of muscle mechanical power output, wing kinematics and unsteady aerodynamic mechanisms. Here, I attempted to draw a broad outline on fluid dynamic mechanisms found in flapping insect wings such as leading edge vorticity, rotational circulation and wake capture momentum transfer, as well as on the constraints of flight force control by the neuromuscular system of the fruit fly Drosophila. This system-level perspective on muscle control and aerodynamic mechanisms is thought to be a fundamental bridge in any attempt to link the function and performance of the various flight components with their particular role for wing motion and aerodynamic control in the behaving animal. Eventually, this research might facilitate the development of man-made biomimetic autonomous micro air vehicles using flapping wing motion for propulsion that are currently under construction by engineers.     

New CNN and hybrid CNN-LSTM models for learning object manipulation of humanoid robots from demonstration

Cluster Computing - As the environments that human live are complex and uncontrolled, the object manipulation with humanoid robots is regarded as one of the most challenging tasks. Learning a...     

Closed-loop learning control of bio-networks.

A general goal of systems biology is to acquire a detailed quantitative understanding of the life-sustaining interactions between genes and proteins. There arises an interesting question of whether these network dynamics can be controlled externally. In the open-loop approach to experimental biology, a control design would be chosen based on a desired target response and modeling with all the available knowledge about the system. If the system is not completely understood or disturbances occur, then unexpected deviations from the desired response can arise. A means to circumvent this difficulty is to optimize the controls in a closed-loop operation by modifying successive input controls based on the performance of previous controls. This paper presents a simulation of closed-loop learning control applied to biological systems in order to generate a desired response. The most powerful advantage of this technique is that the controls are deduced based on experimental results and the process can operate without a model for the underlying biochemical network. This feature eliminates the problem of faulty predictions as well as the need for a detailed understanding of the underlying molecular pathways, suggesting that biological systems can be controlled even before the post-systems biology era.     

Distributed mechanical feedback in arthropods and robots simplifies control of rapid running on challenging terrain

Terrestrial arthropods negotiate demanding terrain more effectively than any search-and-rescue robot. Slow, precise stepping using distributed neural feedback is one strategy for dealing with challenging terrain. Alternatively, arthropods could simplify control on demanding surfaces by rapid running that uses kinetic energy to bridge gaps between footholds. We demonstrate that this is achieved using distributed mechanical feedback, resulting from passive contacts along legs positioned by pre-programmed trajectories favorable to their attachment mechanisms. We used wire-mesh experimental surfaces to determine how a decrease in foothold probability affects speed and stability. Spiders and insects attained high running speeds on simulated terrain with 90% of the surface contact area removed. Cockroaches maintained high speeds even with their tarsi ablated, by generating horizontally oriented leg trajectories. Spiders with more vertically directed leg placement used leg spines, which resulted in more effective distributed contact by interlocking with asperities during leg extension, but collapsing during flexion, preventing entanglement. Ghost crabs, which naturally lack leg spines, showed increased mobility on wire mesh after the addition of artificial, collapsible spines. A bioinspired robot, RHex, was redesigned to maximize effective distributed leg contact, by changing leg orientation and adding directional spines. These changes improved RHex's agility on challenging surfaces without adding sensors or changing the control system.     

Motor control and learning in altered dynamic environments

Dynamic perturbations of reaching movements are an important technique for studying motor learning and adaptation. Adaptation to non-contacting, velocity-dependent inertial Coriolis forces generated by arm movements during passive body rotation is very rapid, and when complete the Coriolis forces are no longer sensed. Adaptation to velocity-dependent forces delivered by a robotic manipulandum takes longer and the perturbations continue to be perceived even when adaptation is complete. These differences reflect adaptive self-calibration of motor control versus learning the behavior of an external object or 'tool'. Velocity-dependent inertial Coriolis forces also arise in everyday behavior during voluntary turn and reach movements but because of anticipatory feedforward motor compensations do not affect movement accuracy despite being larger than the velocity-dependent forces typically used in experimental studies. Progress has been made in understanding: the common features that determine adaptive responses to velocity-dependent perturbations of jaw and limb movements; the transfer of adaptation to mechanical perturbations across different contact sites on a limb; and the parcellation and separate representation of the static and dynamic components of multiforce perturbations.     

Modeling of control and learning in a stepping motion

In a previous study (Beuter et al. 1986) the authors modeled a stepping motion using a three-body linkage with four degrees of freedom. Stepping was simulated by using three task parameters (i.e., step height, length, and duration) and sinusoidal joint angular velocity profiles. The results supported the concept of a hierarchical control structure with open-loop control during normal operation. In this study we refine the dynamic model and improve the simulation technique by incorporating the dynamics of the leg after landing, adding a foot segment to the model, and preprogramming the complete step motion using cycloids. The equations of the forces and torques developed on the ground by the foot during the landing phase are derived using the Lagrangian method. Simulation results are compared to experimental data collected on a subject stepping four times over an obstacle using a Selspot motion analysis system. A hierarchical control model that incorporates a learning process is proposed. The model allows an efficient combination of open and closed loop control strategies and involves hardwired movement segments. We also test the hypothesis of cycloidal velocity profiles in the joint programs against experimental data using a novel curve-fitting procedure based on analytical rather than numerical differentiation. The results suggest multiob-jective optimization of the joint's motion. The control and learning model proposed here will help the understanding of the mechanisms responsible for assembling selected movement segments into goaldirected movement sequences in humans.