>  RESEARCH ACHIEVEMENTS 

> PATH PLANNING OF NON-HOLONMIC ROBOTS
My first research works, during my master degree were on path planning and movement generation
for a non holonomic car like robot. They have allowed me developing good skills in mobile robotics and to deal with the common problems of path planning such as obstacle detection, space exploration, search of optimal trajectories and then to link this with the non-holonomic constraints typical to car like robots, in order to determine the admissible trajectories and the movement the robot is able to achieve. The research has been concluded with the creation of a movement generation simulation software for car like robots.

> IDENTIFICATION OF CAR DYNAMICS
During my PhD I have concentrated my work on the possibility to use robotic formalisms and methods
to identify the car dynamics. It was requested to overcome both principles of vehicle dynamics and more particularly its behavior on the road, and the robotics theory of modelling and identification.
For this work the dynamic behavior of the car has been modelled as a multibody system in order to use all the possibilities that are offered by such a modelling to obtain the dynamic model of the vehicle, which is linear with respect to the dynamic parameters to estimate. Dynamic parameters can then be identified using a solving method of linear over determinate systems. The method chosen for its simplicity of use and availability in calculation software (Matlab, Scilab) is based on the least squares, but a weighted is added to improve the results.
The method has been tested first with simulation software of the car behavior on the road, and then
with a real car equipped with sensors to make the required measurements. Measurements are thus postprocessed in order to make the data usable for the estimation (which procedure is sensitive to noises). Trajectories were set by the car manufacturer to be the common tests. Even sensors were to be among the common sensors that equipped a prototype. The constraint for the identification is the use of four dynamometric wheels to measure the contact forces between the wheels and the ground.
Simulation has allowed validating the model and the method and also validating the available movements on the real car. It has also permitted to define some exciting movements to improve the estimation. Those movements may be added to common trajectories.
Those researches have been carried out with the support of the French car manufacturer PSA Peugeot
Citroen and the Institute of Communication and Cybernetic of Nantes (France). 70% of my PhD have
been spent in the industry in order to be closed to the technical competencies on the vehicle. It has been concluded by the realization of an estimation software for vehicle that is in the toolbox for prototype car tuning engineers.
The modeling and identificationmethod extended to the car dynamics have been published in renowned international conferences in specialized journals.

> FORCE FEEDBACK CONTROL OF A MICROMANIPULATOR
In order to develop much more skills in robotics and control I have completed a post-doctoral fellowship in the Robotics and Intelligent systems service at the Commissariat a l?fEnergie atomique (France). This research work aimed to tele-operate a micro-manipulator. It meant elaborate the control of a piezoresistive cantilever (the manipulator) attached to a nano-translator while manipulated by an haptic arm. The movements of the nano-translator are controlled by an operator via a haptic interface with force feed-back. The force applied by the AFM cantilever is transmitted to the operator. A visual feedback is also available with video-cameras. This work was a good challenge to master the teleoperation skills but also the problems that are linked with different scale worlds between the operator and the objects he has to manipulate (pollen, cells, MEMS components?E). In order to guaranty the overall stability of the system and to allow the operator to manipulate easily the micro and nano objects a coupling scheme based on passive consideration rather than transparency has been chosen. The implementation of the proposed passive coupling scheme was successful to teleoperate the system and to feel clearly the microstage and the nanostage effects while manipulating pollen grains. Several subjects tested the system and even unexperienced subjects manage to grasp and release by rolling a pollen grain. Those researches were done in a collaborative framework between the Commissariat a l?fEnergie Atomique of Fontenay aux Roses (French Nuclear Agency) and the Laboratoire de Robotique de Paris (France).
The proposed control method illustrated by the experimental results have been published in renowned international conferences in a specialized journal.

> IDENTIFICATION OF HUMAN AND HUMANOID DYNAMICS
The human body is a complex and fascinating system; and I have great interests in its understanding in order to design humanoid robots that get closer to humans. The human body is able of sensing, analyzing, deciding and moving thanks to the nervous system and the musculoskeletal system. It is fast, accurate, well-coordinate, compact and powerful, capable of amazing achievements. Impressive examples can be violin prodigy, surgeon, or fencing master; and no humanoid robot can yet compete with. Nevertheless when the body is injured, ageing or suffering from neuro-motor disease such as Parkinsonism, movement disorders and cognitive disorders appear. As such disorders threaten the quality of life of both patient and relatives they are not only an important medical issue, but also a social issue, more particularly in the countries were population is ageing. In that context the study of the human body leading to a sharpened knowledge of the movements generation and control is expressly required to first find medical solutions; second support people by providing personalized robotic home-care. In one hand, many researches focus on the modelling of the musculo-skeletal system to understand and simulate the human movements. They are based on anatomical musculoskeletal geometric descriptions of the human body, and thanks to enhancements of computation power accurate models can now be used. Those models are commonly used to compute kinematics and dynamics such as joint angles and joint torques from the recorded position of optical markers used by motion capture systems. Fields of applications are wide: sport science, rehabilitation, virtual reality, computer aided animation, video games... In addition the dynamics of the human musculo-tendon complex is widely studied by biomechanics and medical researchers. Models are developed to understand the behavior of the muscle as a contractile element that is capable of producing force and changing length when required by the central nervous system. However these models are empirical and based on the Nobel prized Hill model that describes the muscle activation by the neural system. They represent chemical and microscopic aspects of the muscle dynamics to a macroscopic scope. Characteristic individual parameters are not well-known. In the other hand humanoid robots are now able to achieve more domestic tasks, to have a better understanding of their environment, to learn from human. However walk motions in unknown-uneven spaces, grasping, communication, safety are still issues that dramatically limit the use of robots outside the research labs.
In 2004 I have started new research works on these topics at the University of Tokyo in the Nakamura
Laboratory as a JSPS postdoctoral fellow; and which I am continuing now as a Project Assistant Professor. My current research works focus on the identification of dynamics to understand how human are moving, how robot can use this information and how can robot acquire knowledge of their own dynamics. When working on the human body I put emphasis in developing in-vivo, painless and non-invasive methods that are based on the capture of movements and a musculoskeletal model to compute the inverse kinematics and the dynamics; and eventually surface EMG data.
I have focused my researches in three complementary directions. The first is to consider the identification of the inertial parameters of legged systems. The developed method allow to identify the pure inertial parameters, without contamination of inaccurate transmission and friction models. It can be used for any legged system, hence humans, humanoids, quadruped... We are using it to identify the human body dynamics based on motion capture data and force-plate measurements. To optimize the excitation I use motions from a Japanese TV-gymnastic program. We also apply it to identify the dynamics of two humanoid robots: a 50 cm-high robot for which we used combination of internal and external sensors, and a human-size humanoid robot for which we use only the on-board sensors. The obtained parameters can be used for Center of Mass or Zero Moment Point based controllers. I am now considering on-line identification for on-time or real-time applications. The second is focusing on the joint dynamics during passive movements since there is a strong demand from the medical specialists to quantify joint dynamics that is involved in the diagnosis of numerous neuromuscular diseases. To obtain reliable medical information I have been working in collaboration with neurologists from the Tokyo University Hospital. The third is to apply non linear identification methods to a model of muscle in order to estimate the muscle dynamics as muscles are the actuators of the human body; and musculoskeletal based approached to design humanoid robots are very promising. The outcome results of each of these research directions have been published in renowned international conferences and are now under consideration for journal publications.

last update 2008-12-05 by g*