Learning Continuous Grasping Function with a Dexterous Hand from Human Demonstrations

Ye, Jianglong; Wang, Jiashun; Huang, Binghao; Qin, Yuzhe; Wang, Xiaolong

doi:10.48550/arxiv.2207.05053

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, less attention has been paid to hand-object interaction representations for high-level generalization with dexterous hands. For point cloud inputs, features such as position, distance, or/and a global feature vector extracted by PointNet [5] describe the relationship between hands and objects [6], [7], [8]; for input with RGB images, a global feature vector extracted from CNN like ResNet [9] is used to capture the relation [10]. Though able to capture the geometry of objects, global features typically have difficulty in generalizing to new data.…”

Section: Introductionmentioning

confidence: 99%

DexRepNet: Learning Dexterous Robotic Grasping Network with Geometric and Spatial Hand-Object Representations

Liu¹,

Chen²,

Sun³

et al. 2023

Preprint

View full text Add to dashboard Cite

Robotic dexterous grasping is a challenging problem due to the high degree of freedom (DoF) and complex contacts of multi-fingered robotic hands. Existing deep reinforcement learning (DRL) based methods leverage human demonstrations to reduce sample complexity due to the high dimensional action space with dexterous grasping. However, less attention has been paid to hand-object interaction representations for high-level generalization. In this paper, we propose a novel geometric and spatial hand-object interaction representation, named DexRep, to capture dynamic object shape features and the spatial relations between hands and objects during grasping. DexRep comprises Occupancy Feature for rough shapes within sensing range by moving hands, Surface Feature for changing hand-object surface distances, and Local-Geo Feature for local geometric surface features most related to potential contacts. Based on the new representation, we propose a dexterous deep reinforcement learning method to learn a generalizable grasping policy DexRepNet. Experimental results show that our method outperforms baselines using existing representations for robotic grasping dramatically both in grasp success rate and convergence speed. It achieves a 93% grasping success rate on seen objects and higher than 80% grasping success rates on diverse objects of unseen categories in both simulation and real-world experiments.

show abstract