Estimation des Intentions de l'utilisateur pour la navigation semi-autonome d'un fauteuil roulant robotique

User Intention Estimation for Semi-Autonomous Navigation of a Robotic Wheelchair

• Auteur / Autrice : Jesús-Arturo Escobedo-Cabello
• Direction : Christian Laugier, Anne Spalanzani
• Type : Thèse de doctorat
• Discipline(s) : Informatique
• Date : Soutenance le 03/10/2014
• Etablissement(s) : Grenoble
• Ecole(s) doctorale(s) : École doctorale Mathématiques, sciences et technologies de l'information, informatique (Grenoble ; 1995-....)
• Partenaire(s) de recherche : Laboratoire : Laboratoire d'informatique de Grenoble (Isère, France ; 2007-....)
• Jury : Président / Présidente : James L Crowley
• Jury : Examinateurs / Examinatrices : David Daney
• Jury : Rapporteur / Rapporteuse : Rachid Alami, Marie Babel

Mots clés

Mots clés contrôlés

Statistique bayésienne

Mots clés libres

Estimation des Intentions de l'utilisateur

Interaction Sociale

Évasion d'obstacles

Navigation Autonome

Réseaux Bayesiens

User Intention Estimation

Social interaction

Obstacles avoidance

Autonomous Navigation

Bayesian Networks

Résumé

L'auteur n'a pas fourni de résumé en français

This thesis focuses on semi-autonomous wheelchair navigation. We aim to design asystem respecting the following constraints.Safety: The system must avoid collisions with objects and specially with humans present in the scene.Usability: People with motor disabilities and elders often have problems using joysticks and other standard control devices. The use of more sophisticated and human-like ways of interacting with the robot must be addressed to improve the acceptance and comfort for the user. It is also considered that the user could just be able to move one finger and so the request of human intervention should be as reduced as possible to accomplish the navigation task.Compliance:} The robot must navigate securely among obstacles while reducing the frustration caused to the user by taking into account his intentions at different levels; final destination, preferred path, speed etc.Respect of social conventions: When moving, the robot may considerably disturb people around it, especially when its behavior is perceived as unsocial. It is thus important to produce socially acceptable motion to reduce disturbances. We will also addresses the issue of determining those places where the robot should be placed in order become part of an interacting group.In this work we propose to estimate the user's intention in order to reduce thenumber of necessary commands to drive a robotic wheelchair and deal withambiguous or inaccurate input interfaces. In this way, the wheelchair can be incharge of some part of the navigation task and alleviate the user involvement.The proposed system takes into account the user intention in terms of the finaldestination and desired speed. At each level, the method tries to favor themost ``reasonable'' action according to the inferred user intention.The user intention problem is approached by using a model of the user based onthe hypothesis that it is possible to learn typical destinations (those wherethe user spends most of his time) and use this information to enhance theestimation of the destination targeted by the user when he is driving therobotic wheelchair.A probabilistic framework is used to model the existent relationship betweenthe intention of the user and the observed command. The main originality of theapproach relies on modeling the user intentions as typical destinations and theuse of this estimation to check the reliability of a user's command to decidehow much preeminence it should be assigned by the shared controller whenmanaging the robot's speed.The proposed shared-control navigation system considers the direction of thecommands given by the user, the obstacles detected by the robot and the inferreddestination to correct the robot's velocity when necessary. This system is basedon the dynamic window approach modified to consider the input given by the user,his intention, the obstacles and the wheelchair's dynamic constraints tocompute the appropriate velocity command.All of the results obtained in this thesis have been implemented and validatedwith experiments, using both real and simulated data. Real data has beenobtained on two different scenarios; one was at INRIA's entry hall and the otherat the experimental apartment GERHOME.