Vision-based Navigation and Manipulation
From Robot Intelligence
Contents
Concept
- As a map representation, we proposed a hybrid map using object-spatial layout-route information.
- Our global localization is based on object recognition and its pose relationship, and the local localization uses 2D-contour matching by 2D laser scanning data.
- Our map representation is like this:
- The Object-based global localization is as follows:
Related papers
- S Park, Sung-Kee Park, "Global localization for mobile robots using reference scan matching," International Journal of Control, Automation and Systems 12 (1), 156-168, 2014.
- S Kim, H Cheong, DH Kim, Sung-Kee Park, "Context-based object recognition for door detection," Advanced Robotics (ICAR), 2011 15th International Conference on, 155-160, 2011.
- S Park, Sung-Kee Park, "pectral scan matching and its application to global localization for mobile robots," Robotics and Automation (ICRA), 2010 IEEE International Conference on, 1361-1366. 2010.
- S Park, S Kim, M Park, Sung-Kee Park, "Vision-based global localization for mobile robots with hybrid maps of objects and spatial layouts," Information Sciences 179 (24), 4174-4198, 2009.
Unknown Objects Grasping
Concept
- With a stereo vision(passive 3D sensor) and a Jaw-type hand, we studied a method for any unknown object grasping.
- In the context of perception with only one-shot 3D image, three graspable directions such as lift-up, side and frontal direction are suggested, and an affordance-based grasp, handle graspable, is also proposed.
- Our experimental movie clip : https://www.youtube.com/watch?v=YVfTltLy2w0
- Our grasp directions are as follows:
- The schema of our whole grasping process is like this:
