Fig 3 - uploaded by Gioel Asuni
Content may be subject to copyright.
The implementation of the proposed neural model.
Source publication
This paper presents the application of a neural approach in the control of a 7-DOF robotic head. The inverse kinematics problem is addressed, for the control of the gaze fixation point of two cameras mounted on the robotic head. The proposed approach is based on a biologically-inspired model, which replicates the human brain capability of creating...
Context in source publication
Context 1
... is the weight vector associated to the winner unit and N s1 is the set of direct topological neighbors of the winner unit. Respect to the previous implementation, a main variation has been to use only one neural map (Sensory-Motor Map) for coding all the informations of the Motor Position Map, the Spatial Position Map and the Integration Map (see Fig. 3). This modification has permitted a significant reduction in terms of required memory space but above all in terms of computational time. The Sensory-Motor Map has been implemented using the self-organizing growing neural networks (GNG) [13] for correlating the visual perception (gaze fixation point) and proprioception and for their ...
Similar publications
The exploration result on increase in variable drive which creates the impact to
prevail over the electric machine failure due to existence of bearing current. The bearing current
associated with drive system concerning about the operating frequencies of the solid state
semiconductor switches as may cause the electrostatic charge between stator and...
In this paper, the control of DC link capacitor voltages of five-level diode clamped Active Power Filter (APF) using simplified Space Vector Pulse Width Modulation (SVPWM) associated with the redundant vectors of this topology is proposed. The space vector diagram of the five-level inverter is simplified into that of three-level inverter. In turn,...
This study presents a generalised online switching scheme for a space vector pulse-width modulation (SVPWM)-based multilevel inverter of any voltage level. The proposed SVPWM algorithm implements a generalised three-(similar to five) and seven-segment switching scheme using the three most desired switching states and one suitable redundant state fo...
In this paper, a model predictive control (MPC) strategy for a grid-tied cascaded full-bridge inverter with neutral-point-clamped legs has been developed. Two strings of photovoltaic (PV) cells are interfaced to the grid by the inverter while the controller ensures the operation at maximum power conditions. It is thus essential for the developed co...
This study proposes a novel space-vector pulse width modulation scheme for an induction motor with open-end
windings. The proposed PWM scheme requires two two-level inverters with isolated power supplies. This PWM scheme
uses instantaneous phase reference voltages for the generation of gating signals. Open-end winding induction motor
(OEWIM) offers...
Citations
... The proposed method is able to track an object moving in arbitrary trajectories by combining a proportional feedback loop and an adaptive feed-forward gain to predict the next state of the target. Later, Vannucci et al. [175], use an adaptive approach based on Asuni's neural controller [172]. The objective is to coordinate eye and head motions to achieve smooth pursuit of a moving object. ...
... Although gaze direction can, in general terms, be subdivided into the performance of different visual actions, we have decided to set it apart to highlight some works that stand out from the works previously mentioned. We can see, from Table 2, that the proposed solutions in this category are somewhat balanced between intelligent [172,173], predictive or adaptive [167,179] and classical approaches, [186,189], but a new category was also found, that we label Bio-inspired control. ...
... Intelligent control is employed in the works by Asuni et al. [172] and Yoo et al. [173]. Neural networks in the form of Selforganizing Neural Maps (SOM) are used in [172] for the control of the ARTS Lab sensorial head. ...
We conducted a literature review on sensor heads for humanoid robots. A strong case is made on topics involved in human robot interaction. Having found that vision is the most abundant perception system among sensor heads for humanoid robots, we included a review of control techniques for humanoid active vision. We provide historical insight and inform on current robotic head design and applications. Information is chronologically organized whenever possible and exposes trends in control techniques, mechanical design, periodical advances and overall philosophy. We found that there are two main types of humanoid robot heads which we propose to classify as either non-expressive face robot heads or expressive face robot heads. We expose their respective characteristics and provide some ideas on design and vision control considerations for humanoid robot heads involved in human robot interactions.
... In 2014, Vannucci et al. [54] proposed an adaptive controller to realize eye-head coordination with prediction of a moving target. The control model is improved based on the methods in [55], which combine the growing neural gas and the motor babbling technique. These models or methods for the eye-head coordination and their features are summarized in Table IV. ...
Biology can provide biomimetic components and new control principles for robotics. Developing a robot system equipped with bionic eyes is a difficult but exciting task. Researchers have been studying the control mechanisms of bionic eyes for many years and considerable models are available. In this paper, control model and its implementation on robots for bionic eyes are reviewed, which covers saccade, smooth pursuit, vergence, vestibule-ocular reflex (VOR), optokinetic reflex (OKR) and eye-head coordination. What is more, some problems and possible solutions in the field of bionic eyes are discussed and analyzed. This review paper can be used as a guide for researchers to identify potential research problems and solutions of the bionic eyes' motion control.
... In this work we present a model based on a neurocontroller which was first introduced by Asuni and colleagues [11]. They proposed an approach based on the motor babbling technique in conjunction with a growing neural gas to achieve the goal of making a robotic head fixing a static target. ...
... Despite having some remarkable properties, their model had some limitations mainly in terms of performances (it takes around 250 control steps to reach a static target). We propose an adaptive controller, based on [11] able to accomplish a visual pursuit task involving eye-head coordination and prediction of a moving target. In this section we describe the general model of the controller, as depicted in Fig. 1. ...
... The goal of the pursuit controller is to move the current gaze fixation point (gf p) of the robot towards the target point. This controller is an improved version of [11]. The core part of this controller is the Sensory-Motor Map, which is a network able to respond to a stimulus by activating some specific units. ...
Nowadays, increasingly complex robots are being designed. As the complexity of robots increases, traditional methods for robotic control fail, as the problem of finding the appropriate kinematic functions can easily become intractable. For this reason the use of neuro-controllers, controllers based on machine learning methods, has risen at a rapid pace. This kind of controllers are especially useful in the field of humanoid robotics, where it is common for the robot to perform hard tasks in a complex environment. A basic task for a humanoid robot is to visually pursue a target using eye-head coordination. In this work we present an adaptive model based on a neuro-controller for visual pursuit. This model allows the robot to follow a moving target with no delay (zero phase lag) using a predictor of the target motion. The results show that the new controller can reach a target posed at a starting distance of 1.2 meters in less than 100 control steps (1 second) and it can follow a moving target at low to medium frequencies (0.3 to 0.5 Hz) with zero-lag and small position error (less then 4 cm along the main motion axis). The controller also has adaptive capabilities, being able to reach and follow a target even when some joints of the robot are clamped.
... One uses specific hardware, like accelerometers and gyroscope to estimate the 3D posture of the head, and complex control algorithms to compensate the oscillations. The use of inertial information was already proposed by several authors [5], [16], [17]. Typically this kind of techniques is used in binocular robot heads, where gaze is implemented through the coordination of the two eye movements. ...
Visually-guided locomotion is important for au-tonomous robotics. However, there are several difficulties, for instance, the head shaking that results from the robot locomotion itself that constraints stable image acquisition and the possibility to rely on that information to act accordingly. In this article, we propose a controller architecture that is able to generate locomotion for a quadruped robot and to generate head motion able to minimize the head motion induced by locomotion itself. The movement controllers are biologically inspired in the concept of Central Pattern Generators (CPGs). CPGs are modelled based on nonlinear dynamical systems, coupled Hopf oscillators. This approach allows to explicitly specify parameters such as amplitude, offset and frequency of movement and to smoothly modulate the generated oscillations according to changes in these parameters. We take advantage of this particularity and propose a combined approach to generate head movement stabilization on a quadruped robot, using CPGs and a global optimization algorithm. The best set of parameters that generates the head movement are computed by the electromagnetism-like algorithm in order to reduce the head shaking caused by locomotion. Experimental results on a simulated AIBO robot demonstrate that the proposed approach generates head movement that does not eliminate but reduces the one induced by locomotion.
... Outstarlearning [3] is used in the DIRECT model for training the network and modifying the synaptic weights. More recently, Asuni et al. [4] were inspired by this approach and proposed a neural network, also based on outstar learning, that aims to control a robotic head for gazing points in the 3D space. ...
In this paper, we present a spiking neural network architecture that autonomously learns to control a 4 degree-of-freedom robotic arm after an initial period of motor babbling. Its aim is to provide the joint commands that will move the end-effector in a desired spatial direction, given the joint configuration of the arm. The spiking neurons have been simulated according to Izhikevich's model, which exhibits biologically realistic behaviour and yet is computationally efficient. The architecture is a feed-forward network where the input layers encode the intended movement direction of the end-effector in spatial coordinates, as well as the information that is given by proprioception about the current joint angles of the arm. The motor commands are determined by decoding the firing patterns in the output layers. Both excitatory and inhibitory synapses connect the input and output layers, and their initial weights are set to random values. The network learns to map input stimuli to motor commands during a phase of repetitive action-perception cycles, in which Spike Timing-Dependent Plasticity (STDP) strengthens synapses between neurons that are correlated and weakens synapses between uncorrelated ones. The trained spiking neural network has been successfully tested on a kinematic model of the arm of an iCub humanoid robot.
... Several neural-network models have been proposed with the aim of detailing at a neural level the functioning of the circular-reaction principle (e.g. [3], [4], [5], [6], [7], [8]). Notwithstanding the great relevance of these contributions, all the proposed models are limited with respect to one or more aspects of biological plausibility. ...
... sigmoid neurons) with no internal dynamics as those characterizing real neurons (e.g. [3], [4], [5], [6], [7], [8]). The last limitation is the absence of neuroscientific evidence supporting the overall systems' architecture and sensory/motor pattern encoding (e.g. ...
One of the most influential principles of motor development theory, the circular-reaction hypothesis, states that infants perform exploratory movements to acquire efferent-reafferent associations later used to perform goal directed behavior. All models proposed so far to specify this principle lack biological plausibility under some respects. This work proposes a model that starts to overcome these limitations. In particular, the model aims to show that overcoming such limitations in an integrated fashion can shed new light on so-far overlooked phenomena of motor development. This goal is pursed by showing how the model develops biologically plausible connections and movement smoothing mechanisms as emergent outcomes of the interplay between its various biologically-plausible features: a dynamic arm with realistic parameters, an equilibrium-point muscle model, a leaky-neuron neural controller based on population codes, and a Hebb learning rule.
... Several neural-network models have been proposed with the aim of detailing at a neural level the functioning of the circular-reaction principle (e.g. [3], [4], [5], [6], [7], [8], [9], [10], [11]). Notwithstanding the great relevance of these contributions, all these neural models learn to accomplish reaching tasks assuming that there are no obstacles in the environment in which the arm moves and that motor babbling can be used to associate final arm's postures with stimuli. ...
... This allows the system to learn to move in an environment containing obstacles and hence more realistic than the environment considered in previous works where motor babbling was used to acquire reaching capabilities (e.g. [3], [4], [5], [6], [7], [8], [9], [10], [11]). The model is only a first step of a broader research agenda and have many limits in its current implementation: (1) the model might have difficulties in performing linear trajectories when there is no obstacle between the initial hand's posture and the target [27] ; (2) the encoding of joints' proprioception is implemented with neurons having Gaussian activations, while empirical evidence suggests that they should be have a quasi-linear activation in relation to postures' parameters (cf. ...
According to one of the most influential principles of mo-tor development theory, the circular-reaction hypothesis, infants perform exploratory random movements (motor babbling) to acquire efferent-reafferent associations later used to perform goal directed behavior. The models proposed so far to specify this principle learn to accomplish reach-ing tasks by using exploratory movements to associate final arm's pos-tures with stimuli. A limit of these models is that they cannot control the path followed by the hand to arrive to the target and so cannot cope with obstacles. This work proposes a model that starts to overcome this limitation, in particular it proposes a new neural-network architecture that uses motor babbling not to learn stimuli-final postures associations but to learn stimuli-trajectories associations through an Hebb rule. The system controls movement trajectories by regulating the parameters of two Time Based Generators that on their turn generate the sequence of desired positions of the arms' "hand". These types of associations render the system more flexible and capable of coping with obstacles. Prelimi-nary tests run with a 2D kinematic arm demonstrate the viability of the proposed approach.
We developed a biologically plausible control algorithm to move the eyes of a six degrees of freedom robotic head in a human-like manner. Our neurocontroller, written with the neural simulator Nengo, integrates different biological neural models of eye movements, such as microsaccades, saccades, vestibular-ocular reflex, smooth pursuit and vergence. The coordination of the movements depends on the stream of sensory information acquired by two silicon retinas used as eyes and by an inertial measurement unit, which serves as a vestibular system. The eye movements generated by our neurocontroller resemble those of humans when exposed to the same visual input. This robotic platform can be used to investigate the most efficient exploration strategies used to extract salient features from either a static or dynamic visual scene. Future research should focus on technical enhancements and model refinements of the system.
In this work, the authors propose a combined approach based on a controller architecture that is able to generate locomotion for a quadruped robot and a global optimization algorithm to generate head movement stabilization. The movement controllers are biologically inspired in the concept of Central Pattern Generators (CPGs) that are modelled based on nonlinear dynamical systems, coupled Hopf oscillators. This approach allows for explicitly specified parameters such as amplitude, offset and frequency of movement and to smoothly modulate the generated oscillations according to changes in these parameters. The overall idea is to generate head movement opposed to the one induced by locomotion, such that the head remains stabilized. Thus, in order to achieve this desired head movement, it is necessary to appropriately tune the CPG parameters. Three different global optimization algorithms search for this best set of parameters. In order to evaluate the resulting head movement, a fitness function based on the Euclidean norm is investigated. Moreover, a constraint-handling technique based on tournament selection was implemented.
