Where's What? - Towards Autonomous Semantic Mapping of Urban Environments
Speaker: Ingmar Posner , University of Oxford, Department of Engineering ScienceContact:
Date: March 30 2010
Time: 11:00AM to 12:00PM
Location: 32-D463 (Star)
Host: Nick Roy
Nick Roy, email@example.comRelevant URL:
Robotics research in navigation and mapping has reached a maturity which enables the by now routine generation of high-quality large scale metric and topological maps of unstructured environments. With this success, however, comes the realisation that prominent applications in robotics -- such as action selection and human machine interaction -- require information beyond mere metric or topological representations. As a result, researchers throughout the community are becoming increasingly interested in adding higher-order, semantic information to the maps obtained. In this talk we provide a snapshot of ongoing work aiming to enrich standard metric or topological maps as provided by a mobile robot with higher-order semantic information.
The first part of the talk covers our approach to the problem of classifying parts of an urban environment into salient workspace classes using a combination of vision and laser data. Environmental cues are considered for classification at different scales where the first stage considers local classifications only while the second stage incorporates both spatial and temporal contextual information.
The second part of the talk introduces the concept of robot literacy. Our world is densely labeled for human benefit. This human-readable text presents a direct carrier of semantic information which has thus far been neglected in robotics research. We show how this rich semantic information can be leveraged for common robotic tasks such as localisation and planning. Text spotting, the reliable detection and parsing of wild text in images, is at the heart of our system. This remains a nontrivial problem. We present a prototype implementation of an end-to-end pipeline for text spotting in natural scene images suitable for integration into a robotic system. Effectively utilising the detected text requires both linguistic and spatial contextual information. In our system this is provided by a robot with connectivity to the internet. We present results from two applications enabled by this new robotic capability: the semantic interpretation of image content according to arbitrary search terms and text-based localisation.
See other events that are part of
See other events happening in March 2010