In order to enable efficient control of a humanhumanoid in physical contact settings, a real-time solution for a contact observer is required. We propose a novel approach for proprioceptive sensor based contact sensing suitable for affordable personal robots with no force/torque or electric current sensing. We combine robot model knowledge and the output of acceleration resolved quadratic programming wholebody controller to make a prediction of expected position tracking error for computing our proposed contact observer signal. We demonstrate the efficiency of our approach in the experiments of contact detection and estimation of collision direction and intensity on a real humanoid robot Pepper platform controlled by a task-space multi-objective quadratic programming controller.

In this paper, we present developments of a realtime compliant motion control for a personal humanoid robot. Our approach allows to interpret and react to human guidance through touch using only joint encoders measurements to monitor contact direction and intensity on both static and moving links. This novel method is developed with consideration of minimal sensor requirement of the hardware platform to meet high affordability criteria. We demonstrate performances in experiments with a humanoid robot Pepper.

We study the use of humanoid robot technology for physical assistance in motion for a frail person. A careful design of a whole-body controller for a humanoid robot needs to be developed in order to ensure efficient, intuitive and secure interaction between humanoid-assistant and human-patient.Here, we present a design and implementation of a wholebody controller that enables a humanoid robot with a mobile base to autonomously reach a person, perform audiovisual communication of intent, and establish several physical contacts for initiating physical assistance. Our controller uses (i) visual human perception as a feedback for navigation and (ii) joint residual signal based contact detection for closed-loop physical contact creation. We assess the developed controller on a healthy subject and report on the experiments achieved and the results.

For robots to interact with humans in close proximity safely and efficiently, a specialized method to compute whole-body robot posture and plan contact locations is required. In our work, a humanoid robot is used as a caregiver that is performing a physical assistance task. We propose a method for formulating and initializing a non-linear optimization posture generation problem from an intuitive description of the assistance task and the result of a human point cloud processing. The proposed method allows to plan whole-body posture and contact locations on a task-specific surface of a human body, under robot equilibrium, friction cone, torque/joint limits, collision avoidance, and assistance task inherent constraints. The proposed framework can uniformly handle any arbitrary surface generated from point clouds, for autonomously planing the contact locations and interaction forces on potentially moving, movable, and deformable surfaces, which occur in direct physical human-robot interaction. We conclude the paper with examples of posture generation for physical human-robot interaction scenarios.

The object of the Methodology of Experimental Research (M. E. R.) [1][2][3] is to search an optimal strategy which will allow to obtain the largest number of good quality information (i.e. precise and unbiased) concerning a studied phenomenon, while carrying a limited number of experiments.For each problem formulated, the first question which must be studied is the choice of the factors. We must remind that the factors are the parameters that we can control. We must choose the variation limit of these factors, which determines the experimental domain. These variations may have very different orders of magnitude, so that, to be able to compare the factor effects, it is necessary to work with the coded levels of variation for each factor. Each experiment represents a particular point of the experimental domain, and provides a measurement with one or several responses of the phenomenon in this point. We are led to build a matrix each line of which will correspond to an experiment, and each column of which will represent a factor. It is the experimental design. If we try to forecast the studied phenomenon in all the experimental domain, we then choose a mathematic Abstract. In the Methodology of Experimental Research (M.E.R.), the quality of the results depends on the choice of the experimental design. A lot of experimental designs exist. The goal of the presentation is to optimize the choice of an experimental design by neural networks. We present in detail the elaboration of networks which facilitates the choice of an experimental design as a second-degree model which studies six factors in the spherical experimental domain.Key words. Methodology of Experimental Research (M.E.R.) -experimental design -neural networks with non-chaotic dynamics -learning: gradient back propagation.* Correspondence and reprints.