AIC Seminar Series

In this talk, we present the complete details of the architecture and implementation of "Leaving Flatland", an exploratory project that attempts bridge the gap between autonomous perception and action on challenging terrain. The proposed system includes comprehensive localization, mapping, path planning and visualization techniques for a mobile robot to operate autonomously in complex 3D indoor and outdoor environments. We integrate robust Visual Odometry (VO) techniques for localization with real-time 3D mapping methods from stereo data to obtain consistent global models annotated with semantic labels. The system uses octree decomposition of the volume under consideration in a manner so that it can easily scale with the size of the map and is also efficient in terms of required memory requirement. The octree decomposition also means that only the relevant parts of the map are updated, so that the processing time is not dependent on the size of the map. These models can be used by a multi-region motion planner to plan 3D paths to reach a specified goal. The system’s performance is evaluated online on a variety of synthetic data sets as well as in real-time. We also briefly point out future extensions to the architecture to enable it to deal with pose corrections that make it robust to VO errors.

Aravind obtained his M.S. and Ph.D. in Electrical Engineering at the University of Maryland, College Park, USA in 2005 and 2007 and his BE in Electrical Engineering at the Birla Institute of Technology and Science, Pilani, India, in 2000. His research interests are Pattern recognition, Computer vision and Image processing.