Fethi Ben Ouezdou

Dynamic walk of a bipedal robot having flexible feet

The objective of the work is to evaluate the contribution of flexibilities in the feet of a bipedal robot during a dynamic walk. The two major effects are the diminution of the intensity of the normal force when the heel touches the ground and the increase of the phase of double support allowing a better stability of the system. Another notable effect is the analogy between the passive flexion amplitude of the tiptoes and the measured results of human subjects.

Qualitative Adaptive Reward Learning With Success Failure Maps: Applied to Humanoid Robot Walking

In the human brain, rewards are encoded in a flexible and adaptive way after each novel stimulus. Neurons of the orbitofrontal cortex are the key reward structure of the brain. Neurobiological studies show that the anterior cingulate cortex of the brain is primarily responsible for avoiding repeated mistakes. According to vigilance threshold, which denotes the tolerance to risks, we can differentiate between a learning mechanism that takes risks and one that averts risks. The tolerance to risk plays an important role in such a learning mechanism. Results have shown the differences in learning capacity between risk-taking and risk-avert behaviors. These neurological properties provide promising inspirations for robot learning based on rewards. In this paper, we propose a learning mechanism that is able to learn from negative and positive feedback with reward coding adaptively. It is composed of two phases: evaluation and decision making. In the evaluation phase, we use a Kohonen self-organizing map technique to represent success and failure. Decision making is based on an early warning mechanism that enables avoiding repeating past mistakes. The behavior to risk is modulated in order to gain experiences for success and for failure. Success map is learned with adaptive reward that qualifies the learned task in order to optimize the efficiency. Our approach is presented with an implementation on the NAO humanoid robot, controlled by a bioinspired neural controller based on a central pattern generator. The learning system adapts the oscillation frequency and the motor neuron gain in pitch and roll in order to walk on flat and sloped terrain, and to switch between them.

ANALYTICAL AND AUTOMATIC MODELING OF DIGITAL HUMANOIDS

The aim of this article is to achieve a parametric modeling of kinematics, geometrical and inertial properties of the various joints and links which constitute an anthropomorphic biped. The result is the automatic creation of virtual models of humanoid bipeds while respecting intrinsically the inertial and geometrical distribution of each link, according only to two parameters: the total mass and the total height of the system. Future developments are to use the analytical parameters of masses and inertia in analytical dynamic models in order to control humanoids having different masses and heights.

Distributed ground/walking robot interaction

Most of the time, the construction of legged robots is made in an empirical way and the optimization of the mechanical structure is seldom taken into account. In order to avoid spending time and money on the construction of many prototypes to test their performance, a CAD tool and a methodology seem to be necessary. In this way it will be possible to optimize on one hand the kinematic structure of the legs, on the other hand the gaits which will be used by the future robot. Thus, we have developed a methodology to design walking structures such as quadrupeds and bipeds, to simulate their dynamic behavior and analyse their performances. The feet/ground interaction is one of the major problem in the context of dynamic simulation for walking devices. Thus, we focus here about the phenomenon of contact. This paper describes a general model for dynamic simulation of contacts between a walking robot and ground. This model considers a force distribution and uses an analytical form for each force depending only on the known state of the robot system. The simulation includes all phenomena that may occur during the locomotion cycle: impact, transition from impact to contact, contact during support with static friction, transition from static to sliding friction, sliding friction and transition from sliding to static friction. Some examples are presented to show the use of this contact model for the simulation of the foot-ground interaction during a walking gait.

Dynamic walk simulation of various bipeds via ankle trajectory

This paper deals with an approach to carry out the dynamic walk of anthropomorphic bipeds. At first, the way to reproduce a reference walking cycle for a standard biped is proposed. Secondly, the foot/ground interaction model is exposed. A set of elementary transformations is then described to modify the locomotion parameters allowing us to obtain various walking gaits for bipeds having different sizes and weights. Some simulation results are finally given.

Muscle forces prediction of the human hand and forearm system in highly realistic simulation

This paper deals with development of a highly realistic human hand and forearm model (HHF). The adopted model has to be as close as possible to the reality of the human being hand, to address several features, linked to manipulation tasks, grasping objects and daily routine movements like shaving, writing, etc. Also, in order to be realistic, the simulation of the movement is made in real-time; meaning, at the same rate as the natural HHF movement. In this paper we focus on the muscles forces determination for a given task. A relation between the muscles forces and the joints torques is established. The resulting forces, responsible of a dynamical movement of the HHF, are calculated by using an optimization method. The calculation is made during a real-time simulation. The results are discussed and interpreted on an example simulating a basic movement of the HHF. Limitations and further developments of the model are discussed.

Real time implementation of CTRNN and BPTT algorithm to learn on-line biped robot balance: Experiments on the standing posture

This paper describes experimental results regarding the real time implementation of continuous time recurrent neural networks (CTRNN) and the dynamic back-propagation through time (BPTT) algorithm for the on-line learning control laws. Experiments are carried out to control the balance of a biped robot prototype in its standing posture. The neural controller is trained to compensate for external perturbations by controlling the torsos joint motions. Algorithms are embedded in the real time electronic unit of the robot. On-line learning implementations are presented in detail. The results on learning behavior and control performance demonstrate the strength and the efficiency of the proposed approach.

Experience-based learning mechanism for neural controller adaptation: Application to walking biped robots

Neurobiology studies showed that the role of the anterior cingulate cortex of the brain is primarily responsible for avoiding repeated mistakes. According to vigilance threshold, which denotes the tolerance to risks, we can differentiate between a learning mechanism that takes risks, and one that averts risks. The tolerance to risk plays an important role in such learning mechanism. Results have shown the differences in learning capacity between risk-taking and risk avert behaviors. In this paper, we propose a learning mechanism that is able to learn from negative and positive feedback. It is composed of two phases, evaluation and decision-making phase. In the evaluation phase, we use a Kohonen Self Organizing Map technique to represent success and failure. Decision-making is based on an early warning mechanism that enables to avoid repeating past mistakes. Our approach is presented with an implementation on a simulated planar biped robot, controlled by a reflexive low-level neural controller. The learning system adapts the dynamics and range of a hip sensor neuron of the controller in order for the robot to walk on flat or sloped terrain. Results show that success and failure maps can learn better with a threshold that is more tolerant to risk. This gives rise to robustness to the controller even in the presence of slope variations.

Simulation and design of 3-DOF eye mechanism using Listing’s law

This paper presents the simulation of an eye mechanism by implementing Listingpsilas law. Based on these results an eye model has been designed to produce all movements a human eye can have. Saccadic and smooth pursuit movements can be achieved by pan and tilt mechanism while roll movements are also important for VOR, vergence and optokinetic reflexes. Eye plant model has been created to produce all eye movements. Realization of listingpsilas law is necessary for saccadic and smooth pursuit movements. By simulation we show that position of insertion points on an eye ball and placement of pulleys on a tendon driven robot eye are important to mechanically implement listingpsilas law. Integration of roll movement is only important if saccadic and smooth pursuit movements obey listingpsilas law, without effected by this roll movement. The paper discusses constrains and simulation results of 3D rotational motion based on the conditions for listingpsilas law.

Potential of hybridization methods to reducing the dimensionality for multispectral biological images

We address the problem of unsupervised band reduction in multispectral imagery. We propose to use a new hybridization of dimensionality reduction method by combining two categories of bands selection method with projection method and apply it to multispectral data. The algorithm employs the concepts of fuzziness and belongingness (Fuzzy K-means) to provide a better and more adaptive clustering process. However, the Fuzzy hybridized algorithm is applicable to medical imagery. A cluster validity function associated with Bezdeks partition coefficient is employed for evaluation of the dimension reductions performance for this multispectral data. Experiments conducted in this paper confirm the feasibility of the new hybridization for multispectral dimensionality reduction and shows the potential of the proposed approach.