Skip to Main content Skip to Navigation

Méthodes probabilistes pour la planification réactive de mouvements

Abstract : Current motion planning techniques are now capable of planning motions for complex mechanisms in cluttered environments. Nevertheless the adaptation of these planners to scenes with both static and mobile obstacles has been limited so far. One of the reasons is the cost to reflect during queries the dynamic changes of the scene into the data structure preprocessed to capture the free configuration space connectivity. Our main contribution is the proposition of a new planner able to deal with static and mobile obstacles. This hybrid planner combines the PRM framework, originally designed for solving multiple-query problems and that we extended for operating in dynamic scenes with improved diffusion techniques designed to solve single queries without requiring any preprocessing. In practice this planner is also based on several lazy evaluation mechanisms to improve the efficiency of the approach. The design of this new planer has led to the main following development: - The proposition of an original architecture for this planner dedicated to changing environments. This architecture include several lazy evaluation mechanisms to make the planner really efficient. - A new diffusion planner to locally reconnect portions of the roadmap broken by mobile obstacles. We call it Dynamic-Domain RRT planner. This new extension of the original RRT method is able to automatically balance refinement of already explored regions and expansion toward new unknown regions. - Two methods to create cyclic roadmaps to initialise the planner. The first one use the assumption that mobile obstacles are confined to some regions of the space to build roadmap specifically to changes of the obstacles positions. The second one is a general method to capture the connectivity but also the homotopical classes of the free configuration space in a condensed data structure. All these methods are implemented within the software platform Move3D developed at LAAS and evaluated on different kind o f complex 3D examples.
Complete list of metadata

Cited literature [87 references]  Display  Hide  Download
Contributor : Catherine Martineau <>
Submitted on : Wednesday, February 1, 2006 - 11:11:43 AM
Last modification on : Friday, January 10, 2020 - 9:08:09 PM
Long-term archiving on: : Saturday, April 3, 2010 - 10:01:50 PM


  • HAL Id : tel-00011515, version 1


Léonard Jaillet. Méthodes probabilistes pour la planification réactive de mouvements. Automatique / Robotique. Université Paul Sabatier - Toulouse III, 2005. Français. ⟨tel-00011515⟩



Record views


Files downloads