[robotics-worldwide] PhD Student position and scholarship "Planning and Acting in Robotics", LAAS/CNRS, Toulouse, France
felix at laas.fr
Mon Mar 4 01:15:17 PST 2013
A PhD Scholarship in the area of
"Planning and Acting in Robotics"
is being offered at
LAAS-CNRS, University of Toulouse, France
for the fall 2013, under the joint supervision of Felix Ingrand and Malik Ghallab.
This subject is at the intersection of AI and robotics. Its objective is to explore and further develop in a systematic way the hierarchization approach to deliberation, from planning of abstract tasks to the triggering concrete commands. A first step is to study planning methods such as Hierarchical Task Network (HTN) and Angelic Hierarchical Planning (AHP), and to extend them towards the needs of deliberate action in robotics. HTNs offer a convenient representation and practical algorithms beyond classical planning assumptions . AHPs follow a novel and promising approach for hierarchical planning with bonding techniques of reachable sets of states . Both approaches need however to be extended for the purpose of this project along several dimensions, among which:
• opportunistic and context dependent refinement techniques for HTN and/or AHP, integrating execution control, supervision and plan revision mechanisms at various levels.
• durative actions with explicit time and temporal constraints between actions and exogenous events,
• uncertainty with different levels of inaccuracy and non-determinism at the different refinement levels,
• continuous and hybrid state variables and action variables along a hierarchy corresponding to the refinement hierarchy.
An important issue in this approach is to clarify what needs to be synthesized on-line in order to deal with new missions and environments, what can be synthesized off-line, parameterized and reused on-line for handling stable robot capabilities. HTN, AHP and other planning techniques can be developed for performing these syntheses.
Another concern to be taken into account is how to specify or acquire the environment and the robots models needed by planning and acting. Ideas such as “compiling down” frequently performed tasks from a computationally expensive process into lower level skills and commands may be considered.
The research topics of the PhD will cover one or several of the above scientific issues of deliberate actions in robotics, with conceptual and theoretical contributions as well as experimental validations. The latter will be deployed on one of the indoor robot platforms of LAAS, in particular on the PR2 robot.
The work will be carried out mainly within the RIS group at LAAS-CNRS. This group is involved with partnerships with several robotics and AI labs in the world. One or several internships abroad during the thesis remain possible.
PhD advisors: Felix Ingrand felix at laas.fr and Malik Ghallab malik at laas.fr
Application (CV, motivation letter, relevant papers/Master thesis in pdf) from students holding or
finishing a master-level degree in computer science should be sent no later than the 15th of April 2013 to felix at laas.fr and malik at laas.fr.
 M. Ghallab, D. Nau, and P. Traverso. Automated planning: theory and practice. Morgan Kaufmann Publishers, Elsevier, 2004
 B. Marthi, S. Russell, and J. Wolfe, Angelic Hierarchical Planning: Optimal and Online Algorithms. Proceedings of the Eighteenth International Conference on Automated Planning and Scheduling (ICAPS 2008), Sydney, 2008.
More information about the robotics-worldwide