«Regular Format By מאת Ido Zelman עידו זלמ Kinematics of octopus arm movements קינמאטיקה של תנועות זרוע ...»
Thesis for the degree חבור לש קבלת התואר
Doctor of Philosophy דוקטור לפילוסופיה
Submitted to the Scientific Council of the מוגש למועצה המדעית של
Weizmann Institute of Science מכו ויצמ למדע
Rehovot, Israel רחובות, ישראל
By מאת Ido Zelman עידו זלמ Kinematics of octopus arm movements קינמאטיקה של תנועות זרוע התמנו Advisor: Prof. Tamar Flash מנחה: פרופ. תמר פלש שבט תשס"ט February 2009 Summary This study is a part of a large scale research project investigating the motor control system of the octopus by referring to kinematic, biomechanical and neural aspects. Here we focus on the kinematics of octopus arm movements and its underlying principles, by processing digital recordings of live movements that are obtained while an octopus is held in a big glass tank. The project is aimed at better understanding the motor control of the octopus arm, and implementing some of the gained knowledge in biologically inspired hyper-redundant manipulators.
Modeling octopus arm movements is performed by a recently developed system (Yekutieli et al. 2007) dedicated to efficiently reconstruct the arm movements in 3D space from a pair of video records taken by two calibrated cameras. The system is based on a manual segmentation of the octopus arm in each frame of a video sequence. This is a time consuming process when a large data set is considered. During this work we have developed an algorithm for the automatic detection of the virtual backbone of the octopus arm in video records (Zelman et al. 2008) based on a recently presented segmentation algorithm (Galun et al. 2005). Utilizing the automatic detection together with the reconstruction system enables efficient reconstructions of a large set of octopus arm movements. Each movement is essentially modeled as a spatio-temporal profile which describes the configuration of the arm in 3D space as a function of time. A large data set of modeled octopus arm movements has been reconstructed from video sequences by the manual detection method. We believe that the automatic method is an essential tool in order to efficiently reconstruct octopus arm movements of different types.
Collaboration was established with the engineering group of Dr. Ian Walker (Clemson University, USA) in order to automatically operate a soft robot manipulator that was developed to have similar properties and behavior to that of the octopus arm (Walker 2000). It has been found that a quasi-static configuration of the octopus arm can be approximated by a compact geometric description to fit the parameters that control the robot. Both dynamic simulations and real-time experiments that were conducted with the robot have demonstrated that a soft manipulator can automatically mimic some of the behaviors of the octopus arm.
ii The non-rigid octopus arm which lacks any well-defined point requires an uncommon geometric representation. We used curvature and torsion surfaces as a unique description of octopus arm movements which has led to a novel analysis of the kinematics of octopus arm behaviors. We found that mathematical procedures allow to decompose the topographic nature of these surfaces into building blocks. These building blocks were clustered into kinematic primitives, which define temporal motor action in 3D space.
Synthetic rules that utilize these primitives were found to characterize stereotypical behaviors of the arm, and arm movements were classified into sub-groups according to the rules they match. Our findings both suggest the existence of kinematic units as motor primitives used by motor control system of the octopus and give a clearer description of octopus arm behavior. Furthermore, the procedures we have applied establish a novel framework that should be used in the analysis of more octopus arm movements in future research. It may be also of general use by studies referring to biologically inspired hyperredundant robots, or to other biological flexible appendages.
At the end of this work, we present a novel theoretical framework which examines the creation of primitives through an evolutionary process. We found that modular configurations may emerge and be preferred through the process of learning since they allow the octopus arm to easily adapt to different target point in a dynamic environment.
1.1 Main goals of the research
1.2 Thesis outline
2.1 Problems in motor control
2.2 Biological studies of the octopus arm
2.3 Soft robot manipulators
2.4 Motor primitives
3 Modeling octopus data
3.1.1 Experimental setting and system framework
3.1.3 Skeletal representation
3.1.4 3D reconstruction
4 Mimicking octopus behavior by a soft robot manipulator
iv5 Kinematics of octopus arm movements
5.1.1 Spatio-temporal movement as a pair of curvature and torsion surfaces....... 39 5.1.2 Surface decomposition
5.1.2 Clustering algorithm
5.2.1 Extracting surfaces and their units
5.2.2 Clustering motor units
5.2.3 Synthesizing motor actions
5.2.4 Classifying octopus arm movements
5.2.5 Additional results
5.2.6 Characterizing a language of motor primitives
6 Evolution of modular arm configuration
Appendix A Automatic detection results
Appendix B Clustering and classification GMM results
Appendix C Clustering and classifying PCA results
Appendix D Publications based on Ph.D. research
Appendix E Statement about independent collaboration
v Acknowledgements First, and foremost I would like to thank my advisor, Prof. Tamar Flash, for her guidance, patience and help. I owe great thanks to Dr. Yoram Yekutieli and Shlomy Hanassy, both for their work in this subject and for the useful meetings we had together. Thanks to Dr.
Meirav Galun and Dr. Ayelet Akselrod-Ballin who helped to facilitate a part of this work by novel techniques. Thanks to Avi Barliya and Dr. Nadav Kashtan for their friendship and the fruitful discussion we had. Last, I would like to thank my wife Keren and my two daughters Anna & Liya.
vi 1 Introduction The abilities of octopuses to learn, memorize and solve rather complicated behavioral problems have attracted scientists to explore this highly intelligent animal (Wells and Wells, 1957; Fiorito et al. 1990). Octopuses have been observed using the arms for various tasks such as locomotion, food gathering, hunting and sophisticated object manipulation (Mather 1998). The maneuverability of the arms and the ability of the peripheral nervous system to perceive and process information are considered as key attributes of the octopus skills. The efficient nature of the movements is mainly due to the flexible structure of the octopus arm which does not contain any rigid elements.
Structural support and force transmission are achieved through the arm’s musculature – the biomechanical principles governing octopus arm movements differ from those in arms with a rigid skeleton (Kier and Smith 1985; Matzner et al. 2000). These properties and the need to coordinate among eight arms, requires excellent motor control capabilities, indicating that the strategy and principles of the arm motor control are based on unique and novel mechanisms (Sumbre et al. 2001).
The analysis of motion of behaving octopuses, particularly reaching movements (Gutfreund et al. 1996, 1998; Sumbre et al. 2001; Yekutieli et al. 2005a, 2005b) and fetching movements (Sumbre et al. 2001, 2005, 2006), has already led to significant insights. These movements were studied by analyzing the kinematics of specific, meaningful points along the arm which were found to be quite stereotypical.
Electromyographic recordings and detailed biomechanical simulations assisted in revealing common principles which reduce the complexity associated with the motor control of these movements. However, kinematic description of specific points along the arm is insufficient for analyzing octopus arm movements in their full complexity.
1.1 Main goals of the research The main goal of this research is to explore the kinematic characteristics of the movements performed by the octopus arm and to examine kinematic principles that may be used by the octopus. By that we aim to further our understanding of the mechanisms underlying the motor control strategies used in the octopus.
1 Digital recordings of live movements are obtained while an octopus is held in a big glass tank. Our interest in the study of octopus arm movements generally requires the analysis of the shape of the entire arm as it changes while moving in 3D space. Therefore, our first objective was to efficiently process the video records in order to reconstruct the movement in 3D space, and to acquire an essential set of modeled arm movements of different types.
A soft robot has been recently developed to resemble properties of the octopus arm (Jones et al. 2004). It consists of just three sections and it has demonstrated a large repertoire of movements and unique capabilities. The robot is manually controlled, sometimes slowly and roughly, by a trained operator (whose human intuition does not necessarily match the unusual behavior of the octopus arm). We aimed at an alternative operation based on gained biological data, in which the robot can automatically mimic real octopus arm movements without an extensive interaction of an operator.
Motor primitives were recently discussed as possible building blocks of motion, whose combinations can generate a large repertoire of complex motor behaviors (Loeb et al. 2000; Mussa-Ivaldi and Bizzi, 2000; Flash and Hochner, 2005). We have aimed at alternative and novel description of the kinematics of arm movements, by a set of motor primitives that might be stored in memory and used by the octopus motor control system in order to efficiently generate new movements. Based on the knowledge we have gained during the research, we aspire to offer a clearer look on octopus behavior by classifying arm movements into families and sub-families according to the motor primitives they use.
Eventually, by exploring the efficient motion control of the octopus arm, we may also facilitate studies that refer to control of biologically inspired hyper-redundant robots, as well as gain a better understanding of other biological flexible appendages.
1.2 Thesis outline This work is divided into four chapters which are related to different aspects of the kinematics of arm movements performed by the octopus.
Chapter 3 describes the basic step of creating the essential database that is generally required for a study which examines the kinematics or dynamics of octopus arm movements. It details and illustrates the 3D reconstruction process of octopus arm movements from digital recordings. Reconstructed movements are extensively used in the work described by the following chapters. The automatic detection method, which is a significant module in the 3D reconstruction process, is described also in submitted article (Zelman et al. 2008), and diverse results are described in Appendix A. This chapter shortly refers also to a former essential work done in our lab by Yekutieli et al.
In Chapter 4 we show where and how the biology and engineering can meet. We show that a real robot can be automatically operated in a way that mimics octopus behavior. Our results were tested by the group of Ian Walker (Clemson University, USA), where a soft robot was built based on the octopus arm properties.
Chapter 5 describes a significant part of this work, during which an alternative representation for the kinematics of octopus arm movements was explored. Different methods were developed and applied in order to demonstrate the feasibility of movement decomposition into kinematic building blocks. These building blocks, considered as motor primitives, were clustered and used in order to classify and clarify the reconstructed arm movements in our data. Additional results that refer to this work are presented in Appendices B and C.
Finally, Chapter 6 suggests a novel framework which can be used in order to examine how an evolutionary process can influence the kinematics of octopus arm movements. A naïve model of the octopus arm performs in an alternate-goal environment, such that the octopus arm has to reach different target points with its tip. Primary results allow us to hypothesize about a tendency for simple and modular configurations (that may correspond to kinematic primitives) as part of the learning process the octopus undergoes.
3 2 Background This chapter presents a general and short background of motor control definitions, previous octopus studies, design of hyper-redundant manipulators and the notion of motor primitives.
2.1 Problems in motor control The control of robotic systems is generally considered a hard problem, whereby a desired behavior is achieved by deriving the right input parameters for the variables of the system. A common task for a robotic manipulator is the specification of position and orientation for the end-effector at a desired target. To accomplish this task it requires to solve the inverse kinematics problem - the need to translate from a given position and orientation of the end-effector to the geometrical configuration of the whole manipulator.