ROBOT NAVIGATION INCROWDS USING QUEADRILATERAL VELOCITY OBSTACLES (QVO)

This thesis tackles the problem of robot navigation within crowds. We propose a reactive methodology, i.e., such that the robot has no full knowledge of the environment and such that its decisions have to be taken in real time. We rely on a well-known geometric approach, the Velocity Obstacle approa...

Descripción completa

Detalles Bibliográficos
Autor: RICARDO ANDREI RAYA ORTEGA
Tipo de recurso: tesis de maestría
Estado:Versión aceptada para publicación
Fecha de publicación:2017
País:México
Institución:Centro de Investigación en Matemáticas
Repositorio:Repositorio Institucional CIMAT
OAI Identifier:oai:cimat.repositorioinstitucional.mx:1008/871
Acceso en línea:http://cimat.repositorioinstitucional.mx/jspui/handle/1008/871
Access Level:acceso abierto
Palabra clave:info:eu-repo/classification/MSC/OBSTÁCULO A VELOCIDAD
info:eu-repo/classification/cti/1
info:eu-repo/classification/cti/12
info:eu-repo/classification/cti/1203
info:eu-repo/classification/cti/120317
Descripción
Sumario:This thesis tackles the problem of robot navigation within crowds. We propose a reactive methodology, i.e., such that the robot has no full knowledge of the environment and such that its decisions have to be taken in real time. We rely on a well-known geometric approach, the Velocity Obstacle approach. We suppose that we are given a set of trajectory prediction models for a group of mobile obstacles (humans, robots, etc.). The robot takes this information from the environment to go from its initial position to its goal, avoiding collisions. We introduce a conservative discretization of the VOs, which we call Quadrilateral Velocity Obstacle, QVO. It consists of a trapezoid that bounds the VO and is computationally easy to model. We propose two approaches: one Deterministic, and one Probabilistic. The first considers the most probable trajectory for each mobile obstacle. This approach results in an optimization problem with linear constraints in the velocity space. The second approach handles a full trajectory distribution, and decides what velocity to take, both to avoid collisions with obstacles, and to reach its goal. Unlike the deterministic model, this approach contemplates the collision probabilistically. We present results of the two approaches in simulation, for holonomic and non-holonomic agents, and we provide exhaustive evaluations of the different parameters involved in our algorithms.