Multi-objects Grasp Planning In Unstructured Environment

Grasping in unstructured environment,which is considered by Amazon the final engineer-ing challenges found in warehouse automation^[1],is an important capability of robots in the context of future autonomous manipulation.In recent years,the increasing need of warehouse order-fulfillment industry and service industry has created a substantial interest for autonomous robotic solutions in unstructured environment.Both of them require the fast,stable and feasible grasping of various arbitrarily shaped and randomly placed objects.Concerning with the characteristics of multi-objects grasping in unstructured environment,specially the limitation of grasp space caused by various arbitrarily shaping and randomly place-ment,this study proposes new grasp generation and planning methods for grasping in unstruc-tured environment.In unstructured environment,actual grasp configuration space of an object is often limited by object's pose and robot's work space.In order to grasp a randomly placed object in the scene,it is necessary to provide as much feasible grasp configurations as possible for the grasp planning.this study propose a grasp generation method based on skeleton and spherical grid search.First,the object's mean curvature skeleton is chosen as the grasp interest region,and a spherical grid search is used to sampling the grasp configurations around the grasp interest region.Next,the grasp quality function which takes the force-balance and PAC metrics into account is used to filter the unstable grasp configurations,and the remains are stored as the grasp candidates to build the off-line grasp database.In unstructured environment,object interactions such as occlusion,slanting,and stacking can also limit actual grasp configuration space of object,and a unfeasible grasp order may cause scene collapse.This study purpose a grasp planning method based on occlusion grasp configu-ration space in order to plan a safe grasp order and a feasible grasp configuration for each object in the scene.This method unifies different type of object interactions in the unstructured envi-ronment into the occlusion of grasp space,combing with the perceptual information,realizes the inference for the interaction relationship of objects in the scene.Finally,a feasible grasp order and grasp configurations for each objects in the scene are planned.To verify the feasibility of the proposed methods,simulations and experiments are carried out.