PROJECT TITLE :

Learning to Detect Visual Grasp Affordance

ABSTRACT:

Look-primarily based estimation of grasp affordances is desirable when 3-D scans become unreliable due to litter or material properties. We develop a general framework for estimating grasp affordances from two-D sources, including local texture-like measures plus object-class measures that capture previously learned grasp ways. Native approaches to estimating grasp positions are shown to be effective in real-world situations, however are unable to impart object-level biases and will be at risk of false positives. We tend to describe how global cues can be used to compute continuous pose estimates and corresponding grasp point locations, using a max-margin optimization for category-level continuous create regression. We tend to provide a completely unique dataset to evaluate visual grasp affordance estimation; on this dataset we have a tendency to show that a fused methodology outperforms either native or international methods alone, which continuous cause estimation improves over discrete output models. Finally, we demonstrate our autonomous object detection and grasping system on the Willow Garage PR2 robot.