Human Inspired Dexterity in Robotic Manipulation

Japan

Chapter 1

Background: Dexterity in Robotic Manipulation by Imitating Human Beings

Tetsuyou Watanabe    Faculty of Mechanical Engineering, Institute of Science and Engineering, Kanazawa University, Kanazawa, Japan

Abstract

Background for this text is introduced with the basic knowledge for understanding of the contents.

Keywords

Dexterity; Robotic manipulation; Imitation

Contents

1.1Background

1.2Complemental Information

1.2.1Statistically Significant Difference

1.2.2State Space Representation

1.2.3Mechanical Impedance

1.2.4Fundamental Grasping Style

1.2.5Kinematics and Statics of Robots

1.2.6Dynamics of Robots

References

1.1 Background

Human beings have been a kind of textbook for constructing robots because robots perform the tasks that humans do in the support of human beings. However, there are a lot of unrevealed things about human beings. Even if knowledge is given, how to utilize that knowledge during the development of robots is on a case by case basis and there are no general methodologies to effectively transfer the knowledge during the development process. It is impossible to deal with all aspects of human beings. This text is focused on dexterity and aims to understand how human beings acquire dexterity in object manipulation, discusses the possibility of its application in robotic systems, and draws key strategies for dealing with robotic dexterous manipulation in the next generation.

Here, we are focused on up-to-date research about human functions and the method for transferring the functions to robotic manipulation.

Human functions: The level of dexterous manipulation by robots is currently far from that of human beings. What can improve the ability of robots? One hint might be to understand the approach that human beings take in dexterity acquisition. For example, suppose there is a task of grasping and manipulating an unknown object. The shape, softness, weight, and gravitational center are unknown parameters of the presented object. How do human beings identify the physical parameters and use the identified parameters to complete dexterous manipulation? How do human beings learn such procedures? The methodology for the identifying and learning process of human beings could provide valuable insights for the construction of dexterous manipulation of robots. The structure of the human finger and hand play an important role for dexterous manipulation. It was acquired in the process of evolution. The key structures for dexterity are also valid for the key structures of robotic hand design.

Method for transferring human functions to robotic manipulation: It is basically difficult to transfer human manipulation techniques or functions to robotic manipulation, because of the different structures and system architectures. One approach might be imitating or embedding key function/components in robots one by one. Recent attempts to transfer the techniques or functions could give us deep insights in realizing dexterous manipulation of robots. Such attempts could proceed to creating new strategies to design, control, and plan for robotic manipulations, although the functions do not perfectly coincide with the ones for human beings.

This text will consolidate recent approaches from both viewpoints in accelerating the next developments in the dexterous manipulation of robots. To facilitate this understanding, there are two separate sections corresponding to the two viewpoints. The first section focuses on human functions while the second section focuses on transferring the functions to robots. Chapter 2 provides recent revelations in hand anatomy, which lead to human functions for dexterous manipulation. Newly discovered functions give us new viewpoints for constructing robotic manipulation. The concept of muscle synergy has been utilized for controlling robotic hands, but the synergy does not always correspond to that for human or animal evolution. Chapter 3 presents how human beings learn dexterous manipulation in the context of sensorimotor functions. The actual functions of human beings give us different insights to understand dexterous manipulation. Chapter 4 presents a trail of excitation of a multisensory illusion of a surgical robotic system to enhance the dexterity of the control of surgical robotic systems. A hint of the embedded features of human behaviors in the system can be obtained. Chapter 5 examines human reaching behavior when manipulating parallel flexible objects and shows that the optimal hand trajectory is composed of a fifth order polynomial (as in the classic minimum jerk model) and trigonometric terms depending on the natural frequencies of the system and time movement.

The second section provides recent results of design (Chapters 6 and 7), control (Chapters 8 and 9), and planning (Chapter 10) for dexterous robotic manipulation while considering human functions. Chapter 6 presents a novel anthropomorphic robotic hand design, imitating the salient features of the human hand. Chapter 7 presents a novel fingertip design imitating the structure of human fingers, and a robotic hand equipped with the fingertip. Both designs give hints for constructing robotic hands utilizing human hand features. Based on the human function of thumb opposability, Chapter 8 presents a control schema utilizing the concept of passivity. Chapter 9 presents the controller with considering the difference of sensing timing between visual sensors (low sampling rate) and joint sensors (high sampling rate). Chapter 10 presents a planning methodology to manipulate objects with two arms like human beings.

1.2 Complemental Information

To make it easier to understand, complemental information is provided here.

1.2.1 Statistically Significant Difference

When examining the differences between two groups, statistical hypothesis testing is used. Both groups are supposed to be represented by the Student's t-distribution. The testing outputs the P-value (probability value), which indicates the probability for the null hypothesis that each element of the two groups belongs to the same distribution. If the P-value is less than the given level of significance (for example, 0.05 or 5%), the null hypothesis is rejected, and it can be said that there is a statically significant difference between the two groups. This analysis is valid for the two groups. If the number of the target groups is more than two, the post hoc test is performed for the analysis. Table 1.1 shows a summary of which test should be performed in each case. For details, please see text books on statistics, for example [1].

Table 1.1

1.2.2 State Space Representation

State space representation is conducted when modeling a system as a first-order differential equation of the input (u), output (y), and state (x). If the system is linear, the state and observation equations are respectively represented by

   (1.1)

where A, B, C, D are the matrixes. It should be noted that D = 0 for most of the cases because it is the feedthrough term. If the system is nonlinear, the state and observation equations are respectively represented by

   (1.2)

Here, one simple example is shown. The model of mass, damper, and spring is considered and illustrated in Fig. 1.1. Let x be the state, m be the mass, d be the damping coefficient, k be the spring coefficient, and f be the applied force. The equation of motion is then represented by

   (1.3)

Fig. 1.1 Model of mass, damper, and spring.

If letting f be the input, Eq. (1.3) can be formulated as the form of state equation as follows.

   (1.4)

1.2.3 Mechanical Impedance

The concept of mechanical impedance came from the concept of electrical impedance and represents the dynamics of the system by the model of mass, damper, and spring (1.3). If changing Eq. (1.3) as a form of the relationship between force (left side) and motion (right side), the mechanical impedance can be represented by Eq. (1.5). For more details, please see robotic text books such as [2].

   (1.5)

1.2.4 Fundamental Grasping Style

There are several grasping styles, but the most popular and fundamental ones are precision grasp and power grasp, as illustrated in Fig. 1.2. For the precision grasp, only fingertip areas are utilized for grasping, and the mobility of objects is good. An in-hand manipulation can be then realized for this style. For the power grasp, the object is enveloped by several parts of a hand, and the motion of the object is restrained. However, the graspable weight is big, and uncertain disturbances can be easily balanced. For more details, please see text books such as [3].

Fig. 1.2 Fundamental grasping style; (A) precision grasp and (B) power grasp.

1.2.5 Kinematics and Statics of Robots

be the vector representing the position and orientation of the robot at the task frame (m be the joint vector with the n dimension. r is the function of q:

   (1.6)

It should be noted that the problem to derive r from the given q is referred to as forward kinematics while the problem to derive q from the given r is referred to as inverse kinematics. By differentiating Eq. (1.6) with respect to time, we get

   (1.7)

where J(q) be the joint torque vector. From Eq. (1.7) and the principle of virtual work, we have

   (1.8)

For more details, please see robotic text books such as [2].

1.2.6 Dynamics of Robots

The dynamics of robots can be approximately represented by the model of mechanical impedance while accurately derived as the equation of motion. The equation of motion for robotics is given by

   (1.9)

where M is is Coriolis and centrifugal forces, and g(q) is a gravitational term. The equation of motion can be derived by utilizing Lagrange's equations or Newton's and Euler's equations. See robotic text books for a detailed deviation. If considering the w at the task frame, Eq. (1.9) becomes

   (1.10)

where statics term is added.

For more details, please see robotic text books such as [2].

References

[1] Murphy K.R., Myors B., Wolach A.H. Statistical Power Analysis: A Simple and General Model for Traditional and Modern Hypothesis Tests. New York: Routledge; 2009. Available at: https://books.google.co.jp/books/about/Statistical_Power_Analysis.html?id=paYdeZRTsT4C&redir_esc=y.

[2] Yoshikawa T. Foundations of Robotics: Analysis and Control. Cambridge, MA: MIT Press; 1990. Available at: https://mitpress.mit.edu/books/foundations-robotics.

[3] Cutkosky M.R., Howe R.D. Human grasp choice and robotic grasp analysis. In: Dextrous Robot Hands. New York, NY: Springer New York; 1990:5–31. Available at: http://link.springer.com/10.1007/978-1-4613-8974-3_1.

Section 1

Chapter 2

Digital Hand: Interface Between the Robot Hand and Human Hand

Makiko Kouchi; Mitsunori Tada    Human Informatics Research Institute, National Institute of Advanced Industrial Science and Technology, Tokyo, Japan

Abstract

There is a wide gap between robotic and human informatics, and it is difficult to interchange the diverse knowledge accumulated in each research field directly. In this chapter, we review the basics of digital-hand technology: (1) hand anatomy, link structure, and surface mesh; (2) individual and representative hand models; and (3) example cases where the digital-hand model was used in motion and mechanical analysis of the human hand, hoping that digital-hand technology will be an interface between robots and humans; enabling dexterous manipulations in robotics and introducing mechanical and quantitative understanding of grasp in human informatics.

Keywords

Digital hand; Anatomy; Structure; Dimension; Representative model; Individual model; Motion analysis; Motion synthesis

Contents

2.1Introduction

2.2Structure of the Human Hand and Digital Hand

2.2.1Anatomy of the Human Hand

2.2.2Link and Joint of the Digital Hand

2.2.3Surface Mesh of a Digital Hand

2.3Digital-Hand Model With Size Variations

2.3.1Individual Digital-Hand Model

2.3.2Representative Digital Hand Models

2.4Analysis by Digital-Hand Model

2.4.1Motion Analysis and Posture Synthesis

2.4.2Mechanical Analysis

2.5Conclusion

References

2.1