- Fecha: Lunes 10 de diciembre de 2012
- Lugar: Salón de Grados de la ETSI Informática y de Telecomunicación
- Horario: 18:30 h.
- Descripción:
Conferencia impartida por el profesor Nicolas Pugeault, del Center for Vision, Speech and Signal Processing de la Universidad de Surrey, Reino Unido.
Esta conferencia, abierta a todo el público interesado, se imparte con motivo de la inauguración del máster en Ingeniería de Computadores y Redes.- Resumen: This talk will present a novel approach for learning driving behaviour from imitation of a human driver, using a generic, holistic representation of the visual information and a machine learning approach. Visual Gist is a holistic, coarse representation of the visual field that is related to the pre-attentive perception in humans, and has been shown to be a good descriptor for context and environment. In this talk, we show that not only a large amount of driving-relevant information can be detected from visual gist, but also a large portion of the driver’s actions can be predicted.
In a second part, we will extend the learning framework to the more demanding task of learning a steering function directly from visual input and a driver’s actions, and demonstrate autonomous navigation of a robotic platform on a challenging track.
Importantly, the proposed approach is not an engineered the solution but instead uses imitation and machine learning to autonomously identify the relationship between visual features and behaviour, providing an efficient, generic approach to autonomous control.
- Resumen: This talk will present a novel approach for learning driving behaviour from imitation of a human driver, using a generic, holistic representation of the visual information and a machine learning approach. Visual Gist is a holistic, coarse representation of the visual field that is related to the pre-attentive perception in humans, and has been shown to be a good descriptor for context and environment. In this talk, we show that not only a large amount of driving-relevant information can be detected from visual gist, but also a large portion of the driver’s actions can be predicted.
- Organiza: Dpto. Arquitectura y Tecnología de Computadores
- Más información: hector@ugr.es
Inicio /
Conferencia: «Driving me Around the Bend: Learning to Drive From Visual Gist»
Dejar una Respuesta
Usted debe haber iniciado sesión para o publicar un comentario.