A Collision-Aware Cable Grasping Method in Cluttered Environment

TL;DR

A Cable Grasping-Convolutional Neural Network (CG-CNN) designed to facilitate robust cable grasping in cluttered environments is introduced, with commendable success rates, surpassing contemporary state-of-the-art approaches.

Abstract

We introduce a Cable Grasping-Convolutional Neural Network designed to facilitate robust cable grasping in cluttered environments. Utilizing physics simulations, we generate an extensive dataset that mimics the intricacies of cable grasping, factoring in potential collisions between cables and robotic grippers. We employ the Approximate Convex Decomposition technique to dissect the non-convex cable model, with grasp quality autonomously labeled based on simulated grasping attempts. The CG-CNN is refined using this simulated dataset and enhanced through domain randomization techniques. Subsequently, the trained model predicts grasp quality, guiding the optimal grasp pose to the robot controller for execution. Grasping efficacy is assessed across both synthetic and real-world settings. Given our model implicit collision sensitivity, we achieved commendable success rates of 92.3% for known cables and 88.4% for unknown cables, surpassing contemporary state-of-the-art approaches. Supplementary materials can be found at https://leizhang-public.github.io/cg-cnn/ .

Figures (9)

• Figure 1: Grasping cables from cluttered scenes exploiting the grasping samples approach. The grasp candidates are obtained considering the force closure principle. a) CG-CNN estimates grasp quality of grasp candidates. Positive grasp candidates are in green: quality $> 0.5$; Negative grasp candidates are in red : quality $< 0.5$. b) Optimal grasp candidate. c) Physics simulation for grasping.
• Figure 2: Definition of the grasp pose.
• Figure 3: CG-CNN based grasping pipeline (left) and CG-CNN network architecture (right). The grasp sampling method is used to sample grasp candidates of two-jaw gripper from depth image. CG-CNN learns grasp policy $\pi_{\rm CG-CNN}$ by predicting grasp qualities using the cropped depth images of sampled grasp candidates. The optimal grasping pose $g_{\rm{opt}}$ is selected and implemented in simulation and real world based on grasp strategy.
• Figure 4: a) To simulate cable grasping with the accurate model of cables instead of convex hull, we decompose the arbitrary polyhedron of the cable with a set of convex shapes. b) Force closure grasp is shown based on normal direction $\Vec{n}_1,\Vec{n}_2$ and grasp directions $\Vec{g}_1,\Vec{g}_2$ of contact points.
• Figure 5: a) Train loss curve. b) Success rate curve of grasping evaluation in simulation.