Sophie Sakka

Active motor babbling for sensorimotor learning

For a complex autonomous robotic system such as a humanoid robot, motor-babbling-based sensorimotor learning is considered an effective method to develop an internal model of the self-body and the environment autonomously. In this paper, we propose a method of sensorimotor learning and evaluate it performance in active learning. The proposed model is characterized by a function we call the “confidence”, and is a measure of the reliability of state prediction and control. The confidence for the state can be a good measure to bias the next exploration strategy of data sampling, and to direct its attention to areas in the state domain less reliably predicted and controlled. We consider the confidence function to be a first step toward an active behavior design for autonomous environment adaptation. The approach was experimentally validated using the humanoid robot James.

Sensory prediction for autonomous robots

For a complex autonomous robotic system such as a humanoid robot, the learning-based sensory prediction is considered effective to develop a perceptual environment model by itself. We developed a learning system for an autonomous robot to predict the next sensory information from the current sensory information and the expected action. The system we consider contains a learning procedure and a behavior generation procedure. The learning procedure uses a multi layer perceptron minimizing the error between a given sensory input and its predicted value. The behavior generation procedure is based on a uniform probablistic density function to sample the learning data randomly, which is the effective strategy when the system does not have any assumption or knowledge of the environment. We also investigated sensory blind prediction which should allow action plannning as well as offer a reliable forecast for a safe evolution of the robot in the environment. The simulation and experimental results show that the system learns interaction between the robot and the environment in high fidelity.

A generalized 3D inverted pendulum model to represent human normal walking

This paper compares different inverted pendulum models to represent the stance phase of human normal walking. We have developed a model which takes into account the mechanism of the foot during the single support phase, by defining a pivot point under the ground level. Similarly to other models, the pivot point as well as the rod length remain constant during the complete single support phase. Lowering the position of the pivot point allows reducing the vertical amplitude of the center of mass (CoM) trajectory and therefore approaching the real CoM trajectory. We have measured the whole body kinematics of a representative healthy male subject and set a reference CoM trajectory based on multi-body modeling of the human body (16 segments). Then, we have determined a common mathematical definition of two inverted pendulum models extended in the three dimensional space: the classical IP-3D and our GIP-3D model.

Support changes during online human motion imitation by a humanoid robot using task specification

This paper presents a method based on inverse kinematics with task specification for online human to humanoid motion imitation. We particularly focus on the problem of lifting and placing feet on the floor during the motion, allowing change of support during stepping or locomotion. The approach avoids the use of motion primitives that limit the robot motions to what had been learned. A direct transposition of movements is generated, allowing the robot to move freely in space as the human model does, at a velocity close to the reference one. The approach is validated on the humanoid robot NAO and shows very promising results for the use of online motion imitation.

Motion Pattern for the Landing Phase of a Vertical Jump for Humanoid Robots

This paper deals with the generation of motion pattern for humanoid robots vertical jump. The study concentrates on the landing phase of the jump which is the most demanding for the system because of high discontinuity of acceleration at impact time. We first decompose the motion in two functions. The first one starts just before landing and consists in maximizing the value of the discontinuity by reducing the feet/ground velocity. We set the feet, COM and feet/COM velocities according to the supposed height of the jump. At impact time, the second function uses measured data of ground reaction force (GRF) from the ankle force sensors to quickly stop the system while keeping its vertical equilibrium. The influence of each function and their respective parameters is discussed and analyzed, and their validity is tested using the model of an existing humanoid platform

Effects of the Rolling Mechanism of the Stance Foot on the Generalized Inverted Pendulum Definition

This paper addresses the way the rolling mechanism of the human foot influences the Generalized Inverted Pendulum (GIP) model during normal walking. A test on 6 subjects was performed, and two ways of modelling were used: the first one uses the filtered data of the Center of pressure (CoP) trajectory to rebuild the human inverted pendulum; the second one linearizes the CoP trajectory with time, which is equivalent to ignoring the effects of the rolling mechanism of the foot. The results show that the linearized model leads to an observable clear area of convergence of the dynamics support lines showing a neat inverted pendulum, while the non-linearized forces to make the distinction between the three sub-phases of single support: damping, stabilizing, propelling.

Improving skin artifacts compensation for knee flexion/extension and knee internal/external rotation

When the motion of the human body is studied by means of marker-based motion capture systems, one of the main source of error in this analysis is caused by the markers movements with respect to the underlying bones due to skin motion and deformation. In the kinematics estimation of the lower limbs this error is particularly notable for the rotations in the frontal and axial planes. Many algorithms exist trying to compensate this problem, most of them giving satisfactory results in the estimation of the knee flexion/extension, but correcting the effects of the skin deformation in other rotations is still an open issue. We have implemented a new algorithm to compensate this problem. This method was evaluated on a motion of the human lower limbs and compared with another existing global approach. The results of this experiment showed that the proposed algorithm gives much better results in the analysis of human motion, particularly for the internal/external rotation of the knee.