Song, Yaoxian (Westlake University), Cheng, Chun (Fudan University), Fei, Yuejiao (Westlake University), Li, Xiangqing (Westlake University), Liu, Qingchen (Australian National University), YU, Changbin (Australian National University)

2.5D Image-Based Robotic Grasping

Scheduled for presentation during the Regular Session "Learning, Fuzzy and Neural Systems" (FC1), Friday, November 29, 2019, 15:45−17:45, WZ Building Room WZ416

2019 Australian & New Zealand Control Conference (ANZCC), November 27-29, 2019, Auckland, New Zealand

This information is tentative and subject to change. Compiled on April 24, 2024

Abstract

We consider the problem of robotic grasping by 2.5D image data sampling from a real sensor. We design an encoder-decoder neural network to predict grasping policy in real-time which enhances the robustness for the policy generation at different observation heights by fusing depth image and RGB image. We propose an open-loop algorithm to realize robotic grasp operation and evaluate our method in a physical robotic system. The result shows that our method is competitive with the state-of-the-art in grasp performance, real-time and model size. The video is available in https: //youtu.be/Wxw_r5a8qV0.