Thanathorn Phoka

Grasping novel objects with depth segmentation

We consider the task of grasping novel objects and cleaning fairly cluttered tables with many novel objects. Recent successful approaches employ machine learning algorithms to identify points on the scene that the robot should grasp. In this paper, we show that the task can be significantly simplified by using segmentation, especially with depth information. A supervised localization method is employed to select graspable segments. We also propose a shape completion and grasp planner method which takes partial 3D information and plans the most stable grasping strategy. Extensive experiments on our robot demonstrate the effectiveness of our approach.

Regrasp planning for a 4-fingered hand manipulating a polygon

This paper proposes an approach for computing a sequence of finger repositioning that allows a 4-fingered hand to switch from one grasping configuration to another while maintaining a force-closure grasp of a polygon during the entire process. Assuming frictional point contacts, the proposed approach is based on exploring a structure called switching graph. The connectivity of this structure captures ability to switch from one grasp to another and allows regrasp planning to be formulated as a graph search. The proposed approach has been implemented and some preliminary results are presented.

Regrasp planning for a 5-fingered hand manipulating a polyhedron

This paper addresses the problem of a 5-fingered hand manipulating a polyhedron. In particular, assuming frictional point contacts, we present an approach for computing a sequence of finger repositioning that allows the hand to switch from one grasping configuration to another while maintaining a force-closure grasp of the polyhedron during the entire process. The proposed approach captures ability to switch from one grasp to another in a graph structure, allowing regrasp planning to be reduced to a graph search problem. The proposed approach is implemented and preliminary results are presented.

Regrasp Planning of Four-Fingered Hand for Parallel Grasp of a Polygonal Object

This paper proposes a necessary and sufficient condition for parallel grasps. We extend the use of this condition to the task of regrasp planning. In particular, we propose a graph structure called a switching graph which contains information about primitive grasping operations such as finger switching and finger sliding. The problem of regrasp planning is transformed to a graph search problem. Mainly, this work concentrates on a parallel grasp with force closure. Assuming frictional point contacts, the proposed method has been implemented and some preliminary results are presented.

Measurement framework of partial cage quality based on probabilistic motion planning

When an object is caged by a set of fingers, it cannot move arbitrarily far from the caging fingers regardless of what possible rigid motion it takes. Although this condition makes caging an attractive choice for nonprehensile manipulation, directly computing caging configurations is still a complex process. More importantly, using a cage could be too restrictive than necessary in many real cases. This is because some object motions rarely occur in reality. Therefore, some non-caging formations of fingers that only allow these rare escape motions can still effectively be deployed in many caging tasks. This paper introduces a concept of partial cage quality which determine an ability to cage an object of a non-caging formation of fingers, partial cage. We believe that the quality is closely related to possibility of an object escaping from a partial cage. A measurement framework based on motion planning is presented together with experimental results to illustrate and justify the proposed concept.

Regrasp planning of three-fingered hand for a polygonal object

This paper addresses the problem of regrasp planning for a polygon with a large number of edges. We propose an approach for computing sequences of finger repositioning that allow the hand to switch from one grasping configuration to another while maintaining force-closure during the entire process. The proposed approach is based on exploring a structure called switching graph. Complete sets of two-fingered force-closure grasps are computed in grasp space. Adjacent sets of force-closure grasps are merged into a connected set which allows finger repositioning by continuous movements of fingers on adjacent polygonal edges. We present an output sensitive algorithm to construct a switching graph from the obtained connected sets. A method for finding the optimal solution of a finger switching is also presented. The proposed approach has been implemented and some preliminary results are presented.

Contact point clustering approach for 5-fingered regrasp planning

We propose a heuristic approach for a regrasp planning problem. The input with a large number of discrete contact points is considered. In this setting, traditional methods of complete solution is not available. Based on wrench space information of the input, our proposed algorithm clusters the input into groups and chooses a representative contact point from each group. A global graph structure for regrasp planning is then constructed using all force closure grasps that can be formed only by representative contact points. Also described are approaches for finding a regrasping sequence from an arbitrary grasp to a grasp in the global structure. Once such regrasping sequences are found for linking the input initial and target grasps to the global graph structure, the regrasp planning problem can be solved as a graph search. The results from preliminary experiments indicate that our method can solve many problem instances efficiently.

Geometric Reformulation of 3-Fingered Force-Closure Condition

This paper addresses the problem of testing whether three contact points form a 3-fingered force-closure grasp in two dimensions. In particular, assuming frictional point contacts, we present a new necessary and sufficient condition for three fingers to form a force-closure grasp. The proposed condition is based on a technique for representing a friction cone as a line segment in a dual plane. This representation allows force-closure test to be formulated as the problem of intersection detection between a line segment and a convex polygon. The resulting geometric condition is presented along with an efficient algorithm for using the condition in force-closure test.

Measurement framework of partial cage quality

When an object is caged by a set of fingers, it cannot move arbitrarily far from the caging fingers regardless of what possible rigid motion it takes. Although this condition makes caging an attractive choice for nonprehensile manipulation, it could be too restrictive than necessary in many real cases. This is because some object motions rarely occur in reality. Therefore, some non-caging formations of fingers that only allow these escape motions can still effectively be deployed in many caging tasks. This paper proposes a simple framework for measuring the quality of non-caging formation. The framework takes geometrical information of partial cage as inputs and estimates the order of inputs quality which is determined by the effort needed to make an object escape from partial cage. Results from preliminary experiments are presented to help support the proposal of the framework.

Planning optimal independent contact regions for two-fingered force-closure grasp of a polygon

This paper addresses the problem of optimizing the maximal independent contact region for two-fingered force- closure grasp of a rigid object in 2D. Existing methods for optimizing this criterion considered only independent graspable regions on a given pair of edges. We propose an algorithm that takes nearby edges into consideration so that larger independent contact regions can be obtained. Our method takes the input polygons, computes graspable regions for each pair of edges, merge all adjacent regions together, and then find the best independent contact region inscribed in those regions. Two different criteria to define the best independent contact region are studied. The first criterion maximizes area of the axis-parallel rectangle in the configuration space, while the other criterion maximizes the smaller sides length of the rectangle. For a given object with n vertices, the first criterion can be optimized using the algorithm from Karen Daniels et al. in O(n2 log2 n) time, while the other criterion can be optimized using the algorithm from Evanthia Papadopoulou and D. T. Lee in O(n2 log n) time.