Sage Journals: Discover world-class research

Abstract

A grasp planning method based on the volume and flattening of a generalized force ellipsoid is proposed to improve the grasping ability of a dexterous robotic hand. First, according to the general solution of joint torques for a dexterous robotic hand, a grasping indicator for the dexterous hand—the maximum volume of a generalized external force ellipsoid and the minimum volume of a generalized contact internal force ellipsoid during accepted flattening—is proposed. Second, an optimal grasp planning method based on a task is established using the grasping indicator as an objective function. Finally, a simulation analysis and grasping experiment are performed. Results show that when the grasping experiment is conducted with the grasping configuration and positions of contact points optimized using the proposed grasping indicator, the root-mean-square values of the joint torques and contact internal forces of the dexterous hand are at a minimum. The effectiveness of the proposed grasping planning method is thus demonstrated.

Keywords

Dexterous robotic hand grasp planning volume of generalized force ellipsoid flattening of generalized force ellipsoid grasping simulation and experiment

Introduction

The basic task of a dexterous hand is to grasp an object stably. The stability of multifingered grasping is generally characterized by force-closure performance under which the contact forces applied by the fingers can balance arbitrary forces and torques exerted on the object. The target of grasp planning is to find the optimal grasping configuration and positions of contact points. Grasp planning can be classified into grasp planning based on the human hand and grasp planning based on optimization.

Grasp planning based on the human hand

Hong and Tan¹ studied the motion capture of the human hand with the Virtual Programming Language (VPL) data glove and teleoperated the Utah/MIT dexterous hand. Rohling and Hollerbach² captured the positions of fingertips with the SARCOS manipulator and teleoperated the Utah/MIT dexterous hand. Kang and Ikeuchi³ investigated the motion capture of the human hand with a combination of stereo vision and a data glove. However, the capture accuracy and real-time performance are limited when using the motion capture based on data gloves, tactile sensors, and stereo vision. Liu and Zhang⁴ developed a joint space mapping method based on virtual joints and virtual fingers and intuitively verified the mapping in the virtual environment. Fischer et al.⁵ established a relationship between the joint space of the human hand and the Cartesian space of the fingertips with a neural net algorithm, and mapped motion from the human hand to the dexterous hand in the workspace, and finally planned the grasping for the Deutsches Zentrum für Luft-und Raumfahrt e.V. (DLR) hand. Liu et al.⁶ proposed a fingertip mapping method in Cartesian space based on virtual fingers. However, the accuracy of motion mapping can be affected by structural differences between the human hand and dexterous hand. Aleotti and Caselli^7,8 proposed a virtual reality-based Programming by Demonstration system for investigating the grasp recognition in virtual reality and presented a grasp planning algorithm and a grasp synthesis strategy by demonstration. To avoid the difficult 3-D reconstruction and parametric grasp classification, Romero et al.^9,10 presented a human-to-robot grasp mapping system in which the human hand posture was classified according to a single image and the classified posture was mapped to a specific robot hand. However, it is only suitable for volar grasping with tactile feedback and not for the pinch grasp and precision disk grasp owing to visual errors. All of the issues mentioned limit the practical application of grasp planning based on the human hand.

Grasp planning based on optimization

Nakamura et al.¹¹ used the minimum contact internal forces as the grasping indicator and developed a grasp planning under the condition that force-closure constraints and static frictional constraints are satisfied. Markenscoff et al.¹² took the minimum sum of the force/moment applied to the object as a grasping indicator and planned the grasping of polyhedral objects. Zhang et al.¹³ and Li et al.¹⁴ introduced the maximum joint motion isotropy as a grasping indicator and accomplished a grasp planning for three-fingered hand with symmetrical configuration. Ding et al.¹⁵ took the minimum distance between the center of mass of the grasped object and the center of the contact points as a grasping indicator, and planned grasping from the known contact points to the remaining contact points using a nonlinear optimal algorithm. Mo and coworkers^16,17 used the maximum external wrench as a criterion to evaluate the performances of the planning. The grasping indicators mentioned only guarantee the maximum grasping efficiency without describing detailed force mapping relations among the joint torque space, contact force space, and external object force. Li and Sastry¹⁸ and Xiong et al.¹⁹ took the singular value product of the grasping matrix as a grasping indicator to measure the grasping stability and planned the contact points. Singh and Ambike²⁰ used the minimum singular value of the grasping matrix and the norm of the contact internal force as grasping indicators and planned the contact points and grasping forces. Bicchi and Sorrentino²¹ introduced the efficiency index, the ratio of the norm of fingertip output velocity to the norm of the joint input velocity, as a grasping indicator in optimizing the grasping configuration. Guo and Sun²² took the position grade and relative loading capability in the joint space as a grasping indicator in optimizing the grasping configuration. Among the above grasp planning indicators, some consider only the optimization of contact positions without considering the optimization of the grasping configuration, and some consider only the external forces while ignoring the internal forces. Yang and Zhang²³ introduced the concept of the generalized internal force ellipsoid and took the shortest axis of the generalized external force ellipsoid and the longest axis of the generalized internal force ellipsoid as a grasping indicator to accomplish the grasp planning. However, the indicator above cannot reflect the volume and flattening of the generalized external force ellipsoid and the generalized contact internal force ellipsoid simultaneously.

The present article first gives the general solution of joint torques. The generalized external/internal force ellipsoid is then defined, and the volume and flattening of the generalized force ellipsoid are given to reflect the shape and size of the ellipsoid. The article then proposes a new grasping indicator—the maximum volume of a generalized external force ellipsoid and the minimum volume of a generalized contact internal force ellipsoid during accepted flattening. An optimal grasp planning method based on the task is then established. Finally, a grasp planning simulation and grasping experiment are conducted for a three-fingered dexterous hand grasping a sphere.

General solution of joint torque

Figure 1 shows that a dexterous hand grasps an object with the base coordinate system labeled B − XYZ and the object coordinate system labeled o − xyz. Let c_i represent the contact point between the finger and object, and the corresponding coordinate system be labeled $c_{i} - x_{i} y_{i} z_{i}$ . The grasping equations are

G f = w

J^{T} f = τ

where $w = {[F_{x} F_{y} F_{z} M_{x} M_{y} M_{z}]}^{T} \in R^{6}$ is the external wrench; $τ = {[τ_{1} τ_{2} \dots τ_{m}]}^{T} \in R^{m}$ is the joint torque. $G \in R^{6 \times n}$ is the grasping matrix, and $J \in R^{n \times m}$ is the Jacobian matrix. The contact force is $f = {[\begin{matrix} f_{c 1}^{T} & f_{c 2}^{T} & \dots & f_{c k}^{T} \end{matrix}]}^{T} \in R^{n}$ , where k is the number of contact points, and $f_{c i}$ is the contact fsorce at c_i. The contact force at c_i is $f_{c i} = {[f_{c i x} f_{c i y} f_{c i z}]}^{T}$ for the hard finger contact model, and the contact force at c_i is $f_{c i} = {[f_{c i x} f_{c i y} f_{c i z} m_{c i z}]}^{T}$ for the soft finger contact model.

Figure 1.

Grasping an object with a dexterous robotic hand.

Because the dimension of the contact force is larger than that of the external wrench, the grasping matrix, $G \in R^{6 \times n}$ , is a rectangular matrix (n > 6). So, the null space of the grasping matrix, $N (G) \in R^{n - rank (G)}$ , exists certainly. The general solution to equation (1) is

f = G^{+} w + (I - G^{+} G) f_{N}

where G ⁺ is the Moore–Penrose generalized inverse of G , and f _N is the contact internal force. By substituting equation (3) into equation (2), the general solution for the joint torque τ is obtained as

τ = J^{T} G^{+} w + J^{T} (I - G^{+} G) f_{N} .

Establishing the optimal grasping indicator

Volume and flattening of the generalized force ellipsoid

In equation (4), $J^{T} G^{+} w$ and $J^{T} (I - G^{+} G) f_{N}$ are used to balance the external force and contact internal force, respectively, so equation (4) can be divided into two terms

τ_{W} = Q_{W}^{T} w,

τ_{N} = Q_{N}^{T} f_{N} .

where $Q_{W}^{} = {(J^{T} G^{+})}^{T}$ , $Q_{N}^{} = {(J^{T} (I - G^{+} G))}^{T}$ , $Q_{W}^{}$ is the correlation matrix for the external force, and $Q_{N}^{}$ is the correlation matrix for the contact internal force.

If the norm of the joint torque vectors is 1, meaning that $τ^{T} τ = 1$ , τ will be on the surface of the generalized unit ball, and equations (5) and (6) can be rewritten, respectively, as

w^{T} Q_{W} Q_{W}^{T} w = 1

f_{N}^{T} Q_{N} Q_{N}^{T} f_{N} = 1

Equation (7) defines the generalized external force ellipsoid, and all the external forces satisfying equation (7) are on the surface of the generalized external force ellipsoid. Similarly, equation (8) defines the generalized contact internal force ellipsoid, and all the contact internal forces satisfying equation (8) are on the surface of the internal force ellipsoid.

Singular value decomposition is employed to decompose the matrix Q _W, which can be expressed as

Q_{W} = U_{W} [\begin{matrix} Δ_{W} & O \\ O & O \end{matrix}] V_{W}^{T},

where $Δ_{W} = diag [\begin{matrix} δ_{W 1} & δ_{W 2} & \dots & δ_{W r} \end{matrix}]$ . Substituting equation (9) into equation (7), we have

w^{T} U_{W} [\begin{matrix} Δ_{w}^{2} & O \\ O & O \end{matrix}] U_{W}^{T} w = 1 .

The length of the main axis of the generalized external force ellipsoid is $σ_{W i} = 1 / δ_{W i}$ , and the direction of the generalized external force ellipsoid is decided by U_W.

Similarly, the length of the main axis of the generalized contact internal force ellipsoid is $σ_{N i} = 1 / δ_{N i}$ , and the direction of the generalized contact internal force ellipsoid is decided by U _N.

We define the volume v_W and flattening v_W of the generalized external force ellipsoid as

v_{w} = σ_{W 1} σ_{W 2} σ_{W r},

u_{w} = σ_{W min} / σ_{W max} .

The larger the chosen v_w, the larger the generalized external force ellipsoid will be, and the greater its ability to resist the external force will be. The larger the chosen u_w, the closer the generalized external force ellipsoid will be to a generalized ball, and the more uniform its resistance will be to external forces acting from all directions.

We define the volume v_N and flattening u_N of the generalized contact internal force ellipsoid as

v_{N} = σ_{N 1} σ_{N 2} σ_{N r} .

u_{N} = σ_{N min} / σ_{N max},

The smaller the chosen v_N, the smaller the generalized contact internal force ellipsoid will be, and the smaller the contact internal force will be. The larger the chosen u_N, the closer the generalized contact internal force ellipsoid will be to a generalized ball, and the more uniform the contact internal force acting on the object will be.

Maximum volume of the generalized external force ellipsoid and minimum volume of the generalized contact internal force ellipsoid during accepted flattening

The volume and flattening of the generalized force ellipsoid depend on the length and distribution of the main axis of the generalized force ellipsoid. The length of the main axis depends on the matrices Q _W and Q _N, where Q _W and Q _N are decided by J , G , and the contact constraints. J and G are decided by the grasping configuration and positions of contact points. Therefore, when grasping the same object with different grasping configurations and positions of contact points, generalized force ellipsoids with different shapes and sizes are acquired.

When τ is on the surface of the generalized unit ball, the external torque and contact internal force are mapped to the generalized force ellipsoid. We always hope that the generalized external force ellipsoid is as large as possible and its shape is close to a ball as similar as possible, such as it can resist a larger external force in all directions. Meanwhile, we hope that the generalized contact internal force ellipsoid is as small as possible and its shape is close to a ball as similar as possible, and a smaller contact internal force is expected to be on the object to prevent weaker parts of the object from being damaged.

Above all, in considering both the size and shape of the generalized force ellipsoid, the grasping configuration should be chosen properly. In other words, the chosen grasping configuration should maximize the volume of the generalized external force ellipsoid and minimize the volume of the generalized contact internal force ellipsoid during the accepted flattening. The objective function of grasp planning can thus be defined as

{\begin{matrix} \max {v_{W} | u_{W} \geq ε_{W}} \\ \min {v_{N} | u_{N} \geq ε_{N}} \end{matrix}

where ε_W is the accepted flattening of the minimum generalized external force ellipsoid, and ε_N is the accepted flattening of the minimum generalized internal force ellipsoid. ε_W and ε_N are given according to the specific problem. v_W, u_W, v_N, and u_N are the functions of J and G .

Optimal grasp planning method based on a task

The optimal grasp planning method based on a task is as follows.

According to the shape of the object and the structure of the dexterous hand, describe the grasping configuration and positions of contact points with simple parameters, α₁, α₂ ⋯ α_n.

Calculate the grasping matrix $G (α_{1} \dots α_{n})$ of the dexterous hand.

Calculate the Jacobian matrix $J (α_{1} \dots α_{n})$ of the dexterous hand.

According to the contact model, calculate the Moore–Penrose generalized inverse, G ⁺, and then solve $Q_{W} (α_{1} \dots α_{n})$ and $Q_{N} (α_{1} \dots α_{n})$ .

Calculate the length of each main axis of the generalized force ellipsoid, σ_Wi and σ_Ni.

Calculate the grasping indicators, v_W, u_W, v_N, and u_N.

Give the proper ε_W and ε_N, and obtain the feasible ranges $φ = {(α_{1} \dots α_{n}) | u_{W} \geq ε_{W}, u_{N} \geq ε_{N}}$ .

Solve the optimal grasping parameters ϕ_best in the feasible ranges to satisfy the maximum v _W and minimum v_N, thus completing the grasp planning.

Simulation of grasp planning

Figure 2(a) shows a three-fingered dexterous hand grasping a sphere with radius 0.025 m. The object coordinate system O − xyz has an origin at the center of the sphere, and the basic coordinate system B − XYZ has an origin at the center of the equilateral triangle ΔA′B′C′ (with side length c = 0.07 m). The three metacarpophalangeal joints are distributed on the vertices of the equilateral triangle, and local coordinate systems $o_{A'} - x_{A'} y_{A'} z_{A'}$ , $o_{B'} - x_{B'} y_{B'} z_{B'}$ , and $o_{C'} - x_{C'} y_{C'} z_{C'}$ are established on each metacarpophalangeal joint, with Denavit–Hartenberg parameters of the fingers given in Table 1.

Figure 2.

Grasping a sphere with a three-fingered dexterous hand.

Table 1.

Denavit–Hartenberg parameters of finger links.

i	$a_{i - 1}$	$α_{i - 1}$ (°)	d	θ_i (°)	Variable range of joint	Link parameters (m)
1	0	0	l ₁	θ ₁	[−90°, 90°]	l₁ = 0.03 l₂ = 0.072 l₃ = 0.05
2	0	90	0	θ ₂	[0°, 180°]
3	l ₂	0	0	θ ₃	[−150°, 150°]
4	l ₃	0	0	0

The grasping model is simplified as follows:

The center of the sphere is immediately above the center of the equilateral triangle, at a distance l(m), and the matrix for the transformation between the coordinate systems O − xyz and B − XYZ is $R_{O B} = I_{3 \times 3}$ .

The three fingers change position along the edge of a sphere with radius R = 0.025 m. Contact point B is fixed, and its position vector in the object coordinate system O − xyz is $r_{O B} = {[0 R 0]}^{T}$ .

Contact points A and C can be distributed along the edge of the sphere, as shown in Figure 2(b), and their relative positions are determined by α_A and α_C.

The grasping configuration and positions of contact points are completely determined by the three parameters α_A, α_C, and l. The three parameters should thus be planned reasonably to ensure the volume of the generalized external force ellipsoid is maximized and the volume of the generalized contact internal force ellipsoid is minimized.

On the basis of the contact matrices $G (α_{A}, α_{C})$ , G _B, and $G_{C} (α_{A}, α_{C})$ , the grasping matrix of the dexterous hand can be expressed using α_A and α_C as $G (α_{A}, α_{C}) = [G_{A} (α_{A}, α_{C}) G_{C} (α_{A}, α_{C}) G_{B}]$ .

On the basis of the finger Jacobian matrices $J_{A} (α_{A}, α_{C}, l)$ , $J_{B} (l)$ , and $J_{C} (α_{A}, α_{C}, l)$ , the Jacobian matrix of the dexterous hand can be described by $J (α_{A}, α_{C}, l) = diag [\begin{matrix} J_{A} (α_{A}, α_{C}, l) & J_{C} (α_{A}, α_{C}, l) & J_{B} (l) \end{matrix}]$ .

Simultaneously, let $Q_{W} = {(J {(α_{A}, α_{C}, l)}^{T} G {(α_{A}, α_{C})}^{+})}^{T}$ be the correlation matrix for the external force, and let $Q_{N} = {(J {(α_{A}, α_{C}, l)}^{T} (I - G {(α_{A}, α_{C})}^{+} G (α_{A}, α_{C})))}^{T}$ be the correlation matrix for the contact internal force. Substituting the singular values into equations (11) to (14) results in $v_{W} = v_{W} (α_{A}, α_{C}, l)$ , $u_{W} = u_{W} (α_{A}, α_{C}, l)$ , $v_{N} = v_{N} (α_{A}, α_{C}, l)$ , and $u_{N} = u_{N} (α_{A}, α_{C}, l)$ .

The change in the grasping indicator with the parameters α_A, α_C, and l is analyzed as follows.

Fixing l = 0.08 m while changing α_A and α_C

Figure 3 shows how the volume of the generalized external force ellipsoid v_W varies with changes in α_A and α_C. The volume is a maximum in the shaded region, where v_W ≥ 0.07; this shows the largest external grasping force that the object can resist. Figure 4 shows positions of contact points corresponding to the shaded region, where the volume v_Wmax is a maximum at α_A = α_C = 30°.

Figure 3.

Variation of v_W with α_A and α_C.

Figure 4.

Grasping region for maximal v_W.

Figure 5 shows how the flattening of the generalized external force ellipsoid u_W varies with changes in α_A and α_C. The flattening is a maximum in the shaded region, where u _W ≥ 0.4 and the grasping forces are transmitted evenly in all directions. Figure 6 shows the positions of contact points corresponding to the shaded region.

Figure 5.

Variation of u_W with α_A and α_C.

Figure 6.

Grasping region for maximal u_W.

If only considering the grasping indicators of the generalized external force ellipsoid, the region of overlap ( $25 ° \leq α_{A} \leq 35 °$ , $25 ° \leq α_{C} \leq 35 °$ ) between Figures 4 and 6 is the optimal positions of contact points during grasping. In this region, the accepted maximum flattening of the generalized external force ellipsoid, ε _W = 0.4, can be obtained, and the maximum volume of the generalized external force ellipsoid, $\max {v_{W} | u_{W} \geq 0.4} \geq 0.07$ , can be obtained. Therefore, when the dexterous hand grasps a sphere, it is reasonable that contact point A is located in the region $25 ° \leq α_{A} \leq 35 °$ and that contact point C is located in the region $25 ° \leq α_{C} \leq 35 °$ . Thus, the dexterous hand not only resist larger external forces but also has better isotropy to resist external force.

Figure 7 shows how the volume of the generalized contact internal force ellipsoid v_N varies with changes in α_A and α_C. v_N reaches a minimal value in the shaded region, where all values of v_N are equal to or less than 0.014; this is the best situation for grasping. Positions of contact points corresponding to the shaded region are shown in Figure 8.

Figure 7.

Variation of v_N with α_A and α_C.

Figure 8.

Grasping region for minimal v _N: (a) ( $15 ° \leq α_{A} \leq 45 °$ , $15 ° \leq α_{C} \leq 45 °$ ), (b) ( $60 ° \leq α_{A} \leq 90 °$ , $60 ° \leq α_{A} \leq 90 °$ ), (c) ( $15 ° \leq α_{A} \leq 35 °$ , $85 ° \leq α_{A} \leq 90 °$ ), and (d) ( $85 ° \leq α_{A} \leq 90 °$ , $15 ° \leq α_{C} \leq 35 °$ ).

Figure 9 shows how the flattening of the generalized contact internal force ellipsoid u_N varies with changes in α_A and α_C. u_N reaches maximal values in the five shaded regions, where all values of u_N are greater than or equal to 0.7. Figure 10 shows the positions of contact points corresponding to the shaded regions.

Figure 9.

Variation of u_N with α_A and α_C.

Figure 10.

Grasping regions for minimal u_N: (a) ( $5 ° \leq α_{A} \leq 10 °$ , $85 ° \leq α_{C} \leq 90 °$ ), (b) ( $85 ° \leq α_{A} \leq 90 °$ , $5 ° \leq α_{C} \leq 10 °$ ), (c) ( $0 ° \leq α_{A} \leq 10 °$ , $65 ° \leq α_{C} \leq 75 °$ ), (d) ( $65 ° \leq α_{A} \leq 75 °$ , $0 ° \leq α_{C} \leq 10 °$ ), and (e) ( $20 ° \leq α_{A} \leq 45 °$ , $20 ° \leq α_{C} \leq 45 °$ ).

Because of the symmetrical distribution of the three fingers, the positions of contact points are distributed symmetrically not only between Figure 10(a) and (b) but also between Figure 10(c) and (d).

The regions of overlap between Figures 9 and 10 are $20 ° \leq α_{A} \leq 45 °$ and $20 ° \leq α_{C} \leq 45 °$ , where we find minimal ε_N = 0.7 and simultaneously $\min {v_{N} | u_{N} \geq 0.7} \leq 0.014$ . Therefore, if only considering the grasping indicators of the generalized contact internal force ellipsoid, we should rationally select the region $20 ° \leq α_{A} \leq 45 °$ and $20 ° \leq α_{C} \leq 45 °$ for the positions of contact points, which ensures that the contact internal force acting on the object is weak in all directions.

Considering both the maximum volume of a generalized external force ellipsoid and the minimum volume of a generalized contact internal force ellipsoid, the optimal grasping region is $25 ° \leq α_{A} \leq 35 °$ and $25 ° \leq α_{C} \leq 35 °$ , with $α_{A} = α_{C} = 30 °$ being the optimal contact points.

Fixing $α_{A} = α_{C} = 30 °$ while changing l

If α_A and α_C are all 30°, the limited range of l can be obtained as 0.151 m ≥ l ≥ 0.065 m. The two limiting grasping configurations are shown in Figure 11.

Figure 11.

Limiting grasping configurations.

Figure 12 shows how v_w and u_W vary with a change in l. First, v_w has maximum values of 0.18 and 0.19 at the two limit positions l = 0.151 m and l = 0.065 m, respectively. For the configuration l = 0.11 m, v_w has a minimum value of 0.041. Second, u_W decreases with increasing l. Considering the variations of both v_W and u_W with l, both v_W and u_W have maximum values when l is at the limit position l = 0.065 m, where v_w = 0.18 and u_W = 0.5. Therefore, if only considering the grasping indicators of the generalized external force ellipsoid, the optimal grasping configuration is l = 0.065 m.

Figure 12.

Variations of v_W and u_W with l.

Figure 13 shows how v_N and u_N vary with a change in l; both v_N and u_N increase with increasing l. We need to choose l = 0.065 m to minimize v_N but also need to choose l = 0.151 m to maximize u_W, which means that we cannot simultaneously get the minimal v_N and the maximal u_W. However, there is little change in u_N, which ranges from 0.71 to 0.74, over the whole adjustable range of l. Therefore, if only considering the grasping indicator of the generalized contact internal force ellipsoid, the optimal grasping configuration is l = 0.065 m.

Figure 13.

Variations of v_N and u_N with l.

The above simulation results of grasp planning reveal that the optimal grasping configuration and positions of contact points are $α_{A} = α_{C} = 30 °$ and l = 0.065 m.

Grasping experiment

When the drive capability of dexterous hand joints is constant, the grasped object can resist stronger external forces from all directions if the grasping configuration and positions of contact points are optimized effectively. From another viewpoint, we can say that when the external forces acting on the grasped object are constant, the output torque of the dexterous hand joints will be weaker for the optimized grasping configuration and positions of contact points. Therefore, to validate the effectiveness of the grasping indicator proposed in the present article, it is only necessary to experimentally demonstrate that for the same grasped object and constant external forces acting on the object, the output torques of the dexterous hand joints will be weakest with the optimal grasping configuration and positions of contact points.

This section reports a grasping experiment using a three-fingered dexterous hand that grasps a sphere, as shown in Figure 14. The palm of the dexterous hand is placed horizontally on a table, the sphere is centered directly above the palm and has mass of 315 g, and the points of contact between the sphere and the middle finger, thumb, and index finger are denoted A, B, and C, respectively. For a fixed point B and $0.151 m \geq l \geq 0.065 m$ , the experiment investigates the output joint torques of the dexterous hand with different grasping configurations and positions of contact points

Figure 14.

Grasp planning for a sphere.

Fixing l = 0.08 m while changing α_A and α_C

The positions of contact points A and C are changed at intervals of 10° in the experiment. A personal computer reads torque information from joint torque sensors communicating with joint servo control units through a controller area network. Root-mean-square values of the joint torques of the dexterous hand are calculated and given in Table 2.

Table 2.

Distribution of joint torques of grasping for different positions of contact points $\times 10^{- 2} (N \cdot m)$ .

τ _RMS		$α_{A} (^{\circ})$
τ _RMS		10	20	30	40	50	60	70	80	85
$α_{C} (^{\circ})$	10	2.194	2.151	2.127	2.12	2.131	2.16	2.213	2.296	2.416
	20	2.152	2.124	2.112	2.113	2.128	2.158	2.209	2.292	2.426
	30	2.128	2.114	2.104	2.112	2.127	2.156	2.204	2.286	2.427
	40	2.119	2.116	2.114	2.117	2.129	2.153	2.197	2.276	2.421
	50	2.129	2.133	2.129	2.130	2.137	2.154	2.188	2.262	2.423
	60	2.161	2.161	2.157	2.154	2.154	2.159	2.189	2.243	2.431
	70	2.214	2.21	2.206	2.198	2.187	2.18	2.184	2.218	2.421
	80	2.299	2.294	2.289	2.278	2.254	2.234	2.219	2.201	2.542
	85	2.422	2.421	2.421	2.418	2.424	2.428	2.421	2.542	2.664

Figure 15 is a contour map of the root-mean-square value of joint torques versus the position of the contact points. The root-mean-square value of joint torques reaches a minimum of 0.1897 N·m at $α_{A} = α_{C} = 30 °$ , where the contact points are most advantageous to grasping according to the experience of human grasping. Additionally, the root-mean-square value of joint torques and joint torque output increase as α_A and α_C approach 90°.

Figure 15.

Variation of v_W and u_W with l.

Fixing $α_{A} = α_{C} = 30 °$ while changing l

Both α_A and α_C are fixed at 30° while the initial grasping configuration increases gradually from a minimum value of 0.065 m to 0.150 m in the experiment. Statistics of the joint torques are given in Table 3, while statistics of the contact internal forces, obtained by reading six-dimensional fingertip sensors, are given in Table 4.

Table 3.

Distribution of joint torques for different grasping configurations $\times 10^{- 2} (N \cdot m)$ .

l(m)	0.065	0.07	0.075	0.08	0.085	0.09	0.095	0.1	0.105
τ _RMS	1.99	2.03	2.07	2.11	2.14	2.17	2.20	2.23	2.26
l(m)	0.11	0.115	0.12	0.125	0.13	0.135	0.14	0.145	0.15
τ _RMS	2.28	2.31	2.34	2.37	2.39	2.42	2.45	2.48	2.51

Table 4.

Distribution of contact internal forces for different grasping configurations (N).

l(m)	0.065	0.07	0.075	0.08	0.085	0.09	0.095	0.1	0.105
$f_{N_{RMS}}$	0.1229	0.1240	0.1273	0.1292	0.1307	0.1340	0.1359	0.1395	0.1409
l(m)	0.11	0.115	0.12	0.125	0.13	0.135	0.14	0.145	0.15
$f_{N_{RMS}}$	0.1431	0.1460	0.1477	0.1493	0.1522	0.1534	0.1557	0.1580	0.1607

Figure 16 shows how the root-mean-square value of joint torques varies with a change in grasping configuration. The joint torques applied by the dexterous hand increase with increasing l. When the grasping configuration is at the minimum limit (= 0.065 m), the root-mean-square value of the joint torques applied by the dexterous hand is at a minimum.

Figure 16.

Effect of the grasping configuration on joint torques.

Figure 17 shows how the root-mean-square value of contact internal forces varies with a change in grasping configuration. The contact internal forces increase with increasing l. When the grasping configuration is at the minimum limit (= 0.065 m), the root-mean-square value of the contact internal force is at a minimum.

Figure 17.

Effect of the grasping configuration on contact internal forces.

Experimental results

The main findings of the experiments are as follows:

The required joint torques are a minimum for l = 0.065 m and $α_{A} = α_{C} = 30 °$ . The root-mean-square value of joint torques of the dexterous hand is 25.48% less for these parameters than that for l = 0.08 m and $α_{A} = α_{C} = 85 °$ .

The contact internal forces acting on the sphere are a minimum for l = 0.065 m and $α_{A} = α_{C} = 30 °$ . The root-mean-square value of contact internal forces acting on the sphere is 23.51% less for these parameters than for l = 0.15 m and $α_{A} = α_{C} = 30 °$ .

A minimal root-mean-square value of joint torques of the dexterous hand and a minimal root-mean-square value of contact internal forces are obtained when completing the grasping experiment with the grasping configuration and positions of contact points optimized according to “Simulation of grasp planning” section. The effectiveness of the proposed grasping planning method was thus demonstrated.

Conclusions

Using the general solution of joint torque, the present article proposed a grasping indicator—the maximum volume of the generalized external force ellipsoid and the minimum volume of the generalized internal force ellipsoid during accepted flattening—and established an optimal grasp planning method based on the task. Simulation and experimental results showed that the optimal grasping configuration and positions of contact points could be obtained using the proposed grasp planning method.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research is supported by the National Natural Science Foundation of China (51305088) and the Natural Science Foundation of Heilongjiang Province, China (42400621-1-12069-01).

References

Hong

Tan

. Calibrating a VPL dataglove for teleoperating the Utah/MIT hand. In: Proceedings of the IEEE international conference on robotics and automation, Scottsdale, Arizona, USA, 14–19 May 1989, Vol. 3, pp. 1752–1757.

Rolling

Hollerbach

. Optimized fingertip mapping for teleoperation of dexterous robot hands. In: Proceedings of the IEEE international conference on robotics and automation, Atlanta, USA, May 1993, vol. 3, pp. 769–775.

Kang

Ikeuchi

. Toward automatic robot instruction from perception recognizing a grasp from observation. IEEE Trans Robot Autom 1994: 9(4): 432–443.

Liu

Zhang

. Mapping human hand motion to dexterous robotic hand. In: Proceeding of the 2007 IEEE international conference on robotics and biomimetics, Sanya, China, 15–17 December 2007, pp. 829–834.

Fischer

Patrick

VDS

Hirzinger

. Learning techniques in a dataglove based telemanipulation system for the DLR hand. In: Proceedings of the 1998 IEEE international conference on robotics & automation, Leuven, Belgium, 16–20 May 1998, vol. 2, pp. 1603–1608.

Liu

Sun

HX,

Zhang

. Mapping human hand motion to dexterous robotic hands. J Beijing Univ Posts Telecommun 2005; 28(2): 54–58.

Aleotti

Caselli

. Grasp recognition in virtual reality for robot pregrasp planning by demonstration. In: Proceedings of the 2006 IEEE International Conference on Robotics and Automation, Orlando. Sanya, China, 15–18 December 2006, pp. 2801–2806.

Aleotti

Caselli

. Robot grasp synthesis from virtual demonstration and topology-preserving environment reconstruction. In: Proceedings of the 2007 IEEE/RSJ international conference on intelligent robots and systems, San Diego, California, 29 October–2 November 2007, pp. 2692–2697.

Romero

Kjellstrom

Kragic

. Modeling and evaluation of human-to-robot mapping of grasps. In: International conference on advanced robotics, 2009, pp. 1–6.

10.

Romero

Kjellstrom

Kragic

. Human-to-robot mapping of grasps. In: International conference on advanced robotics, 2009, pp. 1–6.

11.

Nakamura

Nagai

Yoshikawa

. Mechanics of coordinative manipulation by multiple robotic mechanisms. In: Proceedings of the IEEE international conference on robotics and automation, Scottsdale, Arizona, USA, 15–19 May 1989, Vol. 8, pp. 44–59.

12.

Markenscoff

Papadimitriou

. The geometry of grasping. Int J Robot Res 1990; 9(1): 61–74.

13.

Zhang

Yun

Sun

. Grasp planning for three-fingered hands with symmetrical configuration. Mach Tool Hydraul 2007; 35(9): 76–78.

14.

Zhang

Guo

. Grasp planning for robotic multi-fingered Hands. TOBOT 2003; 25(5): 409–418.

15.

Ding

Lee

Wang

. Computation of 3-D form-closure grasps. IEEE Trans Robotics and Autom 2001; 17(4): 515–522.

16.

Huang

. Planning of grasping with multifingered hands based on the maximal external wrench. J Mech Eng 2009; 45(3): 258–262.

17.

Lin

. Investigation into the ability of multi-fingered dexterous hand to grasp an explosive. J South China Univ Technol Nat Sci Ed 2015; 43(7): 124–129.

18.

Sastry . Task oriented optimal grasping by multi-fingered robot hands. IEEE Trans Robot Autom 1988; 4(1): 32–48.

19.

Xiong

Dai

Zhang

. Configuration planning of multifingered grasp and its stability index. J Jiang Han Petrol Inst 1997; 19(1): 85–88.

20.

Singh

Ambike

. A soft-contact and wrench based approach to study grasp planning and execution. J Biomech 2015; 48(14): 3961–3967.

21.

Bicchi

Sorrentino

. Dextrous manipulation though rolling. In: Proceedings of the IEEE international conference on robotics and automation, Nagoya, Aichi, Japan, 21–27 May 1995, vol. 3, pp. 452–457.

22.

Guo

Sun

. Multi-objective optimization grasping planning for multifingered robot hand. J Southeast Univ Nat Sci Ed 2012; 42(4): 643–648.

23.

Yang

Zhang

. Optimal grasping planning of multi-fingered dexterous hands based on generalized force ellipsoid. Mech Science Technol 2001; 20(4): 517–519.

Optimal grasp planning for a dexterous robotic hand using the volume of a generalized force ellipsoid during accepted flattening

Abstract

Keywords

Introduction

Grasp planning based on the human hand

Grasp planning based on optimization

General solution of joint torque

Establishing the optimal grasping indicator

Volume and flattening of the generalized force ellipsoid

Maximum volume of the generalized external force ellipsoid and minimum volume of the generalized contact internal force ellipsoid during accepted flattening

Optimal grasp planning method based on a task

Simulation of grasp planning

Fixing l = 0.08 m while changing αA and αC

Fixing α A = α C = 30 ° while changing l

Grasping experiment

Fixing l = 0.08 m while changing αA and αC

Fixing α A = α C = 30 ° while changing l

Experimental results

Conclusions

Footnotes

Declaration of conflicting interests

Funding

References

Fixing l = 0.08 m while changing α_A and α_C

Fixing $α_{A} = α_{C} = 30 °$ while changing l

Fixing l = 0.08 m while changing α_A and α_C

Fixing $α_{A} = α_{C} = 30 °$ while changing l