Ayuda
Ir al contenido

Dialnet


Social navigation of autonomous robots in populated environments

  • Autores: Noé Pérez Higueras
  • Directores de la Tesis: Luis Merino (dir. tes.), Fernando Caballero (codir. tes.)
  • Lectura: En la Universidad Pablo de Olavide ( España ) en 2018
  • Idioma: español
  • Tribunal Calificador de la Tesis: Alberto Sanfeliu Cortés (presid.), Jesús Capitán Fernández (secret.), Brígida Mónica Faria (voc.)
  • Materias:
  • Enlaces
    • Tesis en acceso abierto en:  TESEO  RIO 
  • Resumen
    • Today, more and more mobile robots are coexisting with us in our daily lives.

      As a result, the behavior of robots that share space with humans in dynamic environments is a subject of intense investigation in robotics. Robots must re- spect human social conventions, guarantee the comfort of surrounding people, and maintain the legibility so that humans can understand the robot’s intentions.

      Robots that move in humans’ vicinity should navigate in a socially compliant way; this is called human-aware navigation. These social behaviors are not easy to frame in mathematical expressions. Consequently, motion planners with pre- programmed constraints and hard-coded functions can fail in acquiring proper behaviors related to human-awareness. All in all, it is easier to demonstrate socially acceptable behaviors than mathematically defining them. Therefore, learning these social behaviors from data seems a more principled approach.

      This thesis aims at endowing mobile robots with new social skills for au- tonomous navigation in spaces populated with humans. This work makes use of learning from demonstration (LfD) approaches to solve the problem of human- aware navigation. Different techniques and algorithms are explored and devel- oped in order to transfer social navigation behaviors to a robot by using demon- strations of human experts performing the proposed tasks.

      The contributions of this thesis are in the field of Learning from Demonstra- tion applied to human-aware navigation tasks. First, a LfD technique based on Inverse Reinforcement Learning (IRL) is employed to learn a policy for ”social” local motion planning. Then, a novel learning algorithm combining LfD concepts and sampling-based path planners is presented. Finally, other novel approaches combining different LfD techniques, like deep learning among others, and path planners are investigated. The methods proposed are compared against state- of-the-art approaches and tested in different experiments with the real robots employed in the European projects FROG and TERESA.


Fundación Dialnet

Dialnet Plus

  • Más información sobre Dialnet Plus

Opciones de compartir

Opciones de entorno