[robotics-worldwide] [jobs] Fully funded PhD position at LAAS-CNRS

Daniel Sidobre daniel.sidobre at laas.fr
Mon Jul 27 08:17:08 PDT 2015


Fully Funded PhD Position at LAAS-CNRS in Toulouse, France

We have just obtained a funding for a 3 years PhD from the University 
Paul Sabatier.

The objective is to work on: Temporal models for motions and forces for 
human/robot interactive manipulation.

I suppose the gross salary is about 1684,93 €/month (source: 
http://www.etudiant.gouv.fr/pid20439/contrats-doctoraux.html).

The project selected by  the university can be found here in French: 
http://homepages.laas.fr/daniel/nweb/node/5

Summary of the project:

The objective of this project is to improve substantially the 
interaction between humans and robots for cooperative manipulation and 
object handover. Service robotic becomes a reality, but building robots 
capable to work naturally and safely with humans contact to exchange 
objects, to help disable persons in their daily tasks or to realize 
tasks in contact still requires important development. In particular new 
tools are necessary to better define and control motions.

A large amount of work was done in the domain of interactive robotic, 
but there is not yet robot capable to realize a complete interactive 
task. Consequently, we must understand what are the critical issues. The 
first point, which is outside the scope of this work, is relative to the 
robotic hardware, i.e. building effective hands, variable stiffness 
actuators or integrated sensors. The second concerns the control of the 
numerous axes of these robots (a least 20 for an anthropomorphic hand) 
that didn’t yet receive fully satisfactory solution, but trails towards 
solutions begin to appear like the use of synergies. The planning of 
interactive tasks is a third issue, two approaches are studied, learning 
and planning. Both are complementary, but we are still a long way from 
the solution. We begin also to understand how humans interact and what 
they expect.

We propose to address the two last challenges from an indirect approach 
using new models for motions and mechanical interactions of the robot.

Above these difficulties, it is worth noting that the characteristic and 
the complexity of the interactive tasks come from, on the one hand, the 
complexity of tasks to realize and thus from the difficulty to model 
them and, on the other hand, from the necessary reactivity to adapt the 
robot behavior to the one of its human partner. To describe an 
interactive manipulation task, defining the final state is not enough 
and it is necessary to describe how to achieve it.

As the human partner can move or change its behavior, the robot could be 
required to replan all or part of the task. Given the planning time and 
the complexity of the task, it is necessary to build a hierarchical 
system and to choose command laws that can be better adapted to the 
context. So the switching from one controller to another one must be 
taken into account in order to ensure the stability during transitional 
phases. Using trajectories at the control level appears to be a 
promising approach [Kröger 2012 
<http://dx.doi.org/10.1109/ICRA.2012.6225175>]. It allows to switch 
easily the frames in which motions are defined or to take into account 
the physical constraints (maximum torque and velocity of motors), as 
well as the human safety related to maximum speed and comfort related to 
maximal jerk. One solution is to generate in real time and then to 
execute a trajectory. This trajectory must join up a close point on the 
desired trajectory of the robot from the current situation while 
respecting constraints. The choice of some “close” point of the 
trajectory is a key element, which allow to give priority to the 
temporal definition of the trajectory, or alternatively to the tracking 
accuracy of the associated path. Similarly, the temporal evolution along 
the trajectory at the controller level can be parameterized to accept a 
delay or not. For example, following to an unforeseen event that 
generate a delay along the trajectory, the robot can choose to trigger 
an error, to continue with a delay or to continue compensating the 
delay. These functionalities are presented here for the movement; one 
aspect of the proposed work consists in the extension of these concepts 
to temporal functions describing forces and stiffnesses.

It would be interesting to realize numerous interactive manipulation 
tasks, both for service and industrial cooperative robotic. A study 
carried out among quadriplegic people as part of the ANR-ASSIST 
<http://www.agence-nationale-recherche.fr/?Projet=ANR-07-ROBO-0011> 
project showed that these persons are very interested in recovering 
their privacy. Help with dressing, for example, cannot be described only 
in terms of motions. It is necessary to precise acceptable stiffness, 
position and force limits. We had to deal with the same kinds of 
problems for an industrial collaborative assembly. The description of 
the interactive motions to define the controller includes not only the 
trajectory, but also stiffness and force evolution.

The aim of this work is to define temporal models for interactive 
manipulation tasks and then to show that planners and controllers based 
on these models offer interesting possibilities. These models should 
also help to improve communication between different robot modules, for 
example a planner and a controller.

Experiments and validations will be carried out with the robots of the 
LAAS robots platform <https://www.laas.fr/public/en/robots-platform>, 
such as two PR2 from Willow Garage and Jido equipped with two LWR-4 KUKA 
arms. The team is also involved in different research project related to 
robot interaction (SAPHARI <http://www.saphari.eu/>, SPENCER 
<http://spencer.eu/>, MARDI).



To apply, please send your cover letter and CV to: daniel.sidobre at laas.fr

Looking forward to your applications.
     Daniel.
--
Daniel Sidobre
  Maître de conférences à l'Université Toulouse III - Paul Sabatier
  Chercheur au LAAS-CNRS http://homepages.laas.fr/daniel


More information about the robotics-worldwide mailing list