Autonomous Robotic Manipulation: Real-Time, Deep-Learning Approach for Grasping of Unknown Objects

Figure 9

Grasp generation results: comparison between grasp generated by the GG-CNN with and without RGB-D image preprocessing for shiny and black objects. The upper 2 rows show the generated grasp output without preprocessing of the RGB-D data. Grasps generated are invalid and the depth info is corrupted. The lower 2 rows show the generated grasp output after preprocessing. The depth data was reconstructed leading to valid grasps.