Abstract:In the unstructured real-world, general object grasping with multi-fingered robot hands requires robots to plan stable and reachable grasps for a plethora of unknown objects and be efficient in arbitrary workspace. Motivated by these rigorous demands, we propose a reachability-aware multi-fingered grasp planning framework that can synthesize feasible high-DOF grasps for novel objects in unrestricted workspace. The framework includes an end-to-end voxel-based Grasp Prediction Network (GPN) for generating stable multi-fingered grasp configurations and a learned Grasping Reachability Evaluator (GRE) to estimate kinematic reachability. The GPN takes as input a voxel representation of the scene and directly outputs the predicted collision-free multi-fingered grasps and the associated grasping stability. The GRE then evaluates the grasping reachability of each generated grasp configuration. Based on the predicted grasping reachability and stability, the most feasible grasp is ultimately executed by the robot. In contrast to previous methods that plan successful grasps within restricted workspace solely based on grasping stability, our approach further incorporates a concept of reachability into the online multi-fingered grasp planning via neural networks. The real-world experiments show that our approach outperforms several comparable methods and achieves an average completion rate of 82.2 % for unseen cluttered objects in unrestricted scenarios. All these results indicate that the proposed approach can facilitate general multi-fingered robotic grasping in unstructured environments.

Grasp Planning Based on Scene Grasp Ability in Unstructured Environment

A Cascaded Deep Learning Framework for Real-time and Robust Grasp Planning

Work Day and Night: A Learning Based Illumination Irrelevant Grasp Planning Method

Planning for Grasping Cluttered Objects Based on Obstruction Degree

Active Planning of Robot Navigation for 3D Scene Exploration

Efficient Object Manipulation to an Arbitrary Goal Pose: Learning-based Anytime Prioritized Planning

Grasp Planning Via Hand-Object Geometric Fitting

Grasp Planning Based on Deep Reinforcement Learning: A Brief Survey

Planning Multi-fingered Grasps with Reachability Awareness in Unrestricted Workspace.

3D Eye-to-Hand Coordination for Uninstructed Robot Grasp Planning

Efficient and Robust Robotic Grasping in Cluttered Scenes: A Point Cloud-Based Approach with Heuristic Evaluation.

Regrasp Planning Using Stable Object Poses Supported by Complex Structures

An Efficient and Stable Grasp Planning Method for Stacked Parts

An efficient grasp planning algorithm for 3-D objects considering hand configuration

An Effective Method for Grasp Planning on Objects with Complex Geometry Combining Human Experience and Analytical Approach

Grasp Region Exploration for 7-Dof Robotic Grasping in Cluttered Scenes

Robot Grasp Planning: A Learning from Demonstration-Based Approach

GE-Grasp: Efficient Target-Oriented Grasping in Dense Clutter

Enveloping Grasp Planning of a Three-Fingered Deployable Metamorphic Robotic Grasper.

A System of Robotic Grasping with Experience Acquisition.

DiPGrasp: Parallel Local Searching for Efficient Differentiable Grasp Planning