Faites connaître cette offre !
Reference : UMR5506-ABDKHE-008
Workplace : MONTPELLIER
Date of publication : Wednesday, September 8, 2021
Scientific Responsible name : Abderrahmane Kheddar
Type of Contract : PhD Student contract / Thesis offer
Contract Period : 36 months
Start date of the thesis : 1 November 2021
Proportion of work : Full time
Remuneration : 2 135,00 € gross monthly
Description of the thesis topic
A PhD position is opened at the CNRS (www.cnrs.fr) - University of Montpellier (www.umontpellier.fr) , LIRMM (www.lirmm.fr) on fundamental research in human-humanoid physical interaction with perspectives in assistive daily motions for frail persons.
The main topic of the research is toward humanoid multi-contact planning and control in close physical interaction with a human person (predefined assistive tasks), considering human perception, dynamic balance, safety and high uncertainties that should be considered at the level of both planning and task-space control. The uncertainties that shall be accounted for are for example, human positioning w.r.t the robot, lack of knowledge on actual human strength, uncertainties in human contacts (with its environment) and on robot-human interaction contacts. The work also comprises important efforts toward care-givers skill transfer and trials on real patients in EHPAD centers in South of France.
PROVISIONAL JOINT RESEARCH SCHEDULE
In the first year, the research will focus on modeling the problem of 3D dynamic balance in multi-contact settings from the perspective of fundamental geometric computational aspects (knowledge on polytops, Minkowki sums are appreciated). The idea is to revisit several existing approaches; e.g. those based on the relation between the center of mass and its acceleration and contact forces considering multi-modal fixed/moving contacts (the moving contacts are those used for supporting the motion of the person, the fixed ones are those contacts between the robot and its environment. Comparison to other approaches, e.g. to what extent 3D capture point, extended 3D ZMP can be used in this context, will also be conducted. The first milestone is to optimize the computation time considering a moving contact point. Real experiments will assess the algorithms through a use-case where the humanoid is required to push a rigid box having different weights put on a tilted plan so as to emulate a robot assisting a person to stand-up or to sit, etc. By the end of the year, if possible, we will consider replacing this box with a real healthy person. In order to achieve the experiment with a person, compliant contacts and manipulation control in multi-contact compliant contacts shall be investigated. In parallel to the dynamic balance, the controller shall be extended to handle compliant contacts. As a canonical experience, we will consider the manipulation of a compliant box having a stiffness close to that of a human and various weights.
The second year shall extend dynamic balance studies considering the pair human-robot. That is to say: how the robot can build the human dynamic balance constraint and use it to plan and on-line control of the human-motion assistance. The estimation of the human center of mass has been investigated in several recent studies and could be assessed in various postures. Moreover, the estimation of the contact forces from vision is also investigated at CNRS. We can build on existing technologies and improve them to be adapted to human-humanoid physical interaction with multi-contact settings so as to compute the dynamic balance constraint set of the human while manipulated by the humanoid. This set will be coupled to that of the robot and reconsider the control under the conditions of sudden loss of contact (e.g. when a person stands from a chair, the contact between her/his body and the chair will vanish and hence the dynamic balance constraints will abruptly change the shape). Research shall also consider robustness issues that are related to safety, such as uncertainties in contact locations, uncertainties in contact model, uncertainties in the human center of mass location, etc. During this year, the controller handling compliant multi-contact manipulation is tried on some use-cases defined by HONDA. Experiments on real use-cases will be defined from our current investigations with aging centers (EHPAD) in the South of France. They can be tested with this project using the HONDA robot (Asimo or the newly developed one) on healthy subjects in Japan.
For the last year, depending on previous year achievements, research will be continued but directed toward whole-body control integration for trials with real patients. On the fundamental level, and to increase the safety level of the humanoid assistance, the eventuality for the robot to plan additional contacts on the fly can be considered. That is to say, we consider that the controller is able to generate contacts on demand to increase safety margins. This problem could be approached by the following question: given a subscribed CoM path and an initial region with contact, how and where to generate a contact (eventually on the fly) which conforms to the needed expansion of the CoM dynamic equilibrium region. In other words, what contact to take to enforce a desired CoM path?
Expected outcome for the last year
- Final demonstrator according to HONDA chosen scenarios and requirements (including the location of the demonstrator) in France and in Japan.
- At least one top journal publication
- PhD thesis writing and defense
- Technology recommendations and future plans
- Dissemination and presentation in front of HONDA panel and officials
Starting date: November 2021
Duration: 3 years
Location: mainly in Montpellier (south of France, Mediterranean coast city) with potential short (1 month) or relatively long stays (more than one month in Tokyo, Japan).
Requirements: Master degree in applied mathematics or computational geometry or optimization or theoretical mechanics with good skills in C++, Python programming. Knowledge in IA and machine learning are a plus.
Development environment: https://github.com/jrl-umi3218/mc_rtc
Robotic platforms: HRP-4, Pepper and Honda Asimo (and Honda others)
Constraints and risks
We talk about it on Twitter!