This paper describes the software architecture of a mobile robot which is able to build in real time a structural interpretation of indoor environments using only visual and proprioceptive sensory information. Navigation is guided by this interpretation, improving on classical reactive approaches. We follow a predictive design criterion: the system must anticipate the consequences of its actions, showing predictive understanding of the scene. Specific solutions are given to all perception stages, from low level segment extraction to 3D scene reconstruction based on the current interpretation, including autocalibration of the camera-robot system. This paper focuses in the architecture that integrates all these elements into a high level perception system. A key point is the process of generation, tracking and confirmation of hypothesis which are maintained in a stable internal representation tuned with the agent movements. There is constant interaction between the bottomup perceptive processes, guided by sensory stimuli, and the top-down ones, guided by the previously constructed models.
GeoBot: A High Level Visual Perception Architecture for Autonomous Robots
2006-01-01
1137851 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Autonomous Robots- Visual Perception in Underground Terrains Using Statistical Region Merging
BASE | 2013
|Global Multi-Perspective Perception for Autonomous Mobile Robots
British Library Conference Proceedings | 1995
|Visual Perception Stack for Autonomous Vehicle
Springer Verlag | 2023
|