Apr 3, 2017 - parts, they were later on sprayed with acrylic paint. ... object count and hence the free location on the gripper can be decided to pick new.
Passive Magnetic Latching Mechanisms For Robotic Applications
Thesis by Usman Amin Fiaz
In Partial Fulfillment of the Requirements For the Degree of Masters of Science
King Abdullah University of Science and Technology Thuwal, Kingdom of Saudi Arabia
April, 2017
2
EXAMINATION COMMITTEE PAGE
The thesis of Usman Amin Fiaz is approved by the examination committee
Committee Chairperson: Jeff S. Shamma Committee Members: Bernard S. Ghanem, Taous-Meriem Laleg
3
©April, 2017
Usman Amin Fiaz All Rights Reserved
4
ABSTRACT
Passive Magnetic Latching Mechanisms for Robotic Applications Usman Amin Fiaz This thesis investigates the passive magnetic latching mechanism designs for autonomous aerial grasping and programmable self-assembly. The enormous latching potential of neodymium magnets is a well-established fact when it comes to their ability to interact with ferrous surfaces in particular. The force of attraction or repulsion among the magnets is strong enough to keep the levitation trains, and high speed transportation pods off the rails. But such utilization of these desirable magnetic properties in commercial applications, comes at a cost of high power consumption since the magnets used are usually electromagnets. On the other hand, we explore some useful robotic applications of passive (and hence low cost) magnetic latching; which are of vital importance in autonomous aerial transportation, automated dronebased package deliveries, and programmable self-assembly and self-reconfigurable systems. We propose, and implement a novel, attach/detach mechatronic mechanism, based on passive magnetic latching of permanent magnets for usBots; our indigenously built programmable self-assembly robots, and show that it validates the game theoretic self-assembly algorithms. Another application addressed in this thesis is the utilization of permanent magnets in autonomous aerial grasping for Unmanned Aerial Vehicles (UAVs). We present a novel gripper design for ferrous objects with a passive magnetic pick up and an impulse based drop. For both the applications, we highlight the importance, simplicity and effectiveness of the proposed designs while providing a brief comparison with the other technologies out there.
5
ACKNOWLEDGEMENTS
A huge thanks to my parents who have always been there for me as a source of inspiration behind every success in my life so far. I would like to thank my adviser, Prof. Jeff S. Shamma for his continuous support, and valuable guidance over the course of my research. It would not have been possible without him. I would also like to thank our research scientist, Dr. Jiming Jiang and our administrative assistant Jielin Li, for their unselfish and timely assistance with logistics, and in procurement and purchase of various components. Finally, special mentions for M. Abdelkader, M. Shaqura, Dr. Hassan Jaleel, and Dr. Mohamed A. Mabrok, for their kind help with various research tools that led to the completion of this thesis.
6
TABLE OF CONTENTS
Examination Committee Page
2
Copyright
3
Abstract
4
Acknowledgements
5
List of Abbreviations
8
List of Figures
9
List of Tables
11
1 Introduction 1.1 Objectives and Contributions . . . . . . . . . . . . . . . . . . . . . . 1.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12 12 13
1.2.1 1.2.2
Self-Assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . Aerial Grasping . . . . . . . . . . . . . . . . . . . . . . . . . .
15 18
Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
2 usBot: A Robot for Programmable Self-Assembly 2.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21 21 24
1.3
2.2.1 2.2.2 2.3 2.4
Design Constraints . . . . . . . . . . . . . . . . . . . . . . . . Design & Features . . . . . . . . . . . . . . . . . . . . . . . .
25 25
2.2.3 The Latching Mechanism . . . . . . . . . . . . . . . . . . . . . External Actuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29 31 33
7 3 Passive Aerial Grasping of Ferrous Objects
36
3.1 3.2
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Working Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36 37
3.3
Passive Magnetic Gripper . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Gripper Design . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Gripper Assembly . . . . . . . . . . . . . . . . . . . . . . . . .
38 38 41
3.3.3 3.3.4
Quadrotor and Control System . . . . . . . . . . . . . . . . . Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42 44
Heavy Duty Gripper . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 New Gripper Design . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Gripper Assembly . . . . . . . . . . . . . . . . . . . . . . . . .
50 51 55
3.4.3 The Quadrotor . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.4 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . Autonomous Aerial Grasping using EPM Based Gripper . . . . . . .
56 59 65
3.4
3.5
4 Conclusion 4.1 4.2
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . .
68 68 69
References
70
Appendices
74
8
LIST OF ABBREVIATIONS
EPM
Electro-Permanent Magnet
GPS LED
Global Positioning System Light Emitting Diode
MagLev Magnetic Levitation MRI Magnetic Resonance Imaging RISC Robotics Intelligent Systems & Control UAV UGV
Unmanned Aerial Vehicle Unmanned Ground Vehicle
9
LIST OF FIGURES
1.1 2.1 2.2
Comparison among various classes of magnets. . . . . . . . . . . . . . Representation of a target assembly. . . . . . . . . . . . . . . . . . . . Overview of the game theoretic approach for self-assembly. . . . . . .
14 22 22
2.3 2.4 2.5
Simulation results for Singleton and Linchpin. . . . . . . . . . . . . usBot: A programmable self-assembly robot. . . . . . . . . . . . . . . usBot: top side view and bottom lateral view of the design. . . . . . .
23 24 26
2.6 2.7
usBot: the expanded lateral view of the design. . . . . . . . . . . . . usBot: the enlarged view of the latching mechanism. . . . . . . . . .
27 30
2.8 The external actuation platform for usBot locomotion. . . . . . . . . 2.9 The force test results for usBots. . . . . . . . . . . . . . . . . . . . . 2.10 The torque test results for usBots. . . . . . . . . . . . . . . . . . . . .
32 33 34
2.11 usBots: experiments showing the primary self-assembly moves. . . . . 2.12 usBots: work in progress. . . . . . . . . . . . . . . . . . . . . . . . . .
35 35
3.1 3.2 3.3
Passive gripper design for ferrous objects. . . . . . . . . . . . . . . . . The gripper assembled and mounted on the quadrotor. . . . . . . . . Relative breakdown of mass of the aerial vehicle. . . . . . . . . . . . .
39 42 43
3.4 3.5
Top side view of the quadrotor. . . . . . . . . . . . . . . . . . . . . . Experimental results for maximum payload test. . . . . . . . . . . . .
44 45
3.6 3.7 3.8
Experimental results for minimum payload test. . . . . . . . . . . . . Experimental setup for maximum sustainable slide. . . . . . . . . . . Experimental results for maximum sustainable slide. . . . . . . . . . .
46 47 48
3.9 Aerial grasping experiment indoors. . . . . . . . . . . . . . . . . . . . 3.10 Heavy duty passive gripper design. . . . . . . . . . . . . . . . . . . .
49 52
3.11 Gripper drop mechanism design. . . . . . . . . . . . . . . . . . . . . . 3.12 Gripper drop mechanism for the real system. . . . . . . . . . . . . . . 3.13 The assembled gripper with all components. . . . . . . . . . . . . . .
54 54 55
3.14 The quadrotor with heavy duty gripper. . . . . . . . . . . . . . . . . 3.15 Customized 2D gimbal. . . . . . . . . . . . . . . . . . . . . . . . . . .
57 58
3.16 Experimental results for maximum payload test. . . . . . . . . . . . .
61
10 3.17 Experimental results for minimum payload test. . . . . . . . . . . . .
61
3.18 Autonomous aerial grasping experiment outdoors. . . . . . . . . . . . 3.19 The aerial grasping capabilities of the heavy duty gripper. . . . . . .
64 64
3.20 Electro-Permanent Magnet (EPM) based magnetic gripper. . . . . . . 3.21 Screen-shot of the frames from the gripper camera. . . . . . . . . . .
65 66
11
LIST OF TABLES
2.1
usBot: List of Components . . . . . . . . . . . . . . . . . . . . . . . .
29
12
Chapter 1 Introduction
1.1
Objectives and Contributions
The primary objective is to design and implement intelligent systems by incorporating the impressive latching ability of passive magnets, and to fabricate, high performance prototypes for the specific applications of programmable self-assembly and autonomous aerial grasping. It is desired to have a test-bench for our self-assembly algorithms that have been either proposed earlier or will pop up later in future. The designed systems thus need to be flexible in order to accommodate and adapt to future prospects as well. Also, the intent is to have a robust and agile grasping mechanism for autonomous aerial grasping via UAVs. In light of the pre-described objectives, and the work done over the course of the thesis, the main contributions can be listed as follows: Design and implementation of usBot; a programmable self-assembly robot, in-
digeniously built at Robotics Intelligent Systems & Control (RISC) lab. Novel and intelligent attach/detach mechanism for usBots. Two magnetic gripper designs for passive aerial grasping of ferrous objects. Novel and intelligent drop mechanism for the grippers.
In addition, the thesis proposes new paths for future, regarding the extension of designed systems beyond being just prototypes. It also hints on developing new and
13 more efficient algorithms for the problem of self-assembly and self-reconfiguration for instance. What follows in this chapter is a brief overview of the related work done in the field over the years, with specific emphasis on the applications of interest.
1.2
Related Work
Magnetic latching with its enormous applications has been around for years [1], [2]. From decades ago to even recent years [3], it has been a tempting topic for the people who work either in industry or research, to investigate the useful properties of magnetic substances and invent corresponding applications, where they might come in handy. It is impressive to appreciate that after so many years of cutting edge research and work along the way, there exists still a lot, to be explored in terms of the utilization of magnets in tasks that could be of any industrial and social importance, or of any research interest. There is no comparison for the reliability, and consistency in the way magnets interact with each other as well as with other ferrous objects. Moreover, in the modern world, they exist as different variants; permanent magnets, electromagnets, and EPMs being the three main classes. All these categories have their own strengths and weaknesses, and hence the choice of a specific class is application dependent. For example, both electromagnets and EPMs require some external actuation or control signal for their function. On the other hand, permanent magnets do not need any excitation from the environment to perform, which is a plus. However, the draw back is an obvious one i.e. no direct control on their operation. Fig. 1.1 provides with a quick comparison of the the three main classes of magnets in terms of their key features. However, there can be intelligent methods to use the permanent magnets in a controlled fashion as is detailed in this thesis in a couple of ways. Another promising feature of permanent magnets is their zero power consumption (hence sometimes referred to as passive), whereas
14
Figure 1.1: Comparison among different classes of magnets in terms of their power consumption, scalability, latching strength and control. Green means good, yellow represents satisfactory and red means unsatisfactory characteristics.
the electromagnets, which are widely used for many industrial and socially important applications such as Magnetic Resonance Imaging (MRI) machines, speakers, motors, generators, particle accelerators [4], and fast Magnetic Levitation (MagLev) trains, consume an enormous amount of power. But for heavy duty applications and for high performance, power consumption is something that is normally compromised. For low scale and power efficient systems however, EPMs provide a pretty good and cost effective solution, since they are much more conservative in terms of power consumption, in comparison with the electromagnets, and yet provide some control over the activation/deactivation of the magnets. This control is generally binary or two state (on/off). However, as is explained in detail, in later chapters of the thesis that for certain applications, which are of enormous interest for researchers, roboticists, and industry, some limitations are associated with the operation of EPMs as well; aerial grasping and programmable self-assembly being two of these applications for instance. In such scenarios, we come up with intelligent designs and smart ways to use passive magnets in a controlled fashion to effectively accomplish the task. The
15 following two sub-sections of this chapter provide specific overview of the existing literature and the summary of the related work that has been carried out on the targeted robotic applications.
1.2.1
Self-Assembly
Millions of natural processes around us employ small organisms, that combine or collaborate in a specialized manner to execute a certain task. Self-assembly refers to the origin of patterned structures that emerge from the collective behavior of simpler participating elements acting in an autonomous fashion. The biological processes that involve the concept of self-assembly are well investigated topics of research during the past decade [5], [6]. The intrinsic motivation for this great deal of research is two fold. First, to understand self-assembly in general, to improve our knowledge of the natural self-assembling systems [7], [8]. Second, to design and develop intelligent systems with the potential of artificial self-assembly. In this thesis, we emphasize on developing systems that would demonstrate the potential for programmable selfassembly. Though the interest in this topic goes back to early 90’s, only recent years have shown some encouraging results regarding the possibility of artificial swarm and collective behavior of multi-agent systems as an implementation [9]. The applications and prospects for programmable self-assembly could be enormous. Some applications are already out there like self-assembling modular structures and the standard units with swappable payloads [10]. These find useful applications in medical industry as well for example self-reconfigurable wheel chairs for the handicapped etc. Re-configurable furniture [11] is another social implication of the programmable self-assembly in particular for the old, and the sick. Self-assembling robots can be very effective in disaster stricken areas where there can not be any human or conventional Unmanned Ground Vehicle (UGV) intervention. These modular micro scale robots with their ability to re-configure themselves in any desired
16 configuration might make it possible to perform future search and rescue or repair missions autonomously. The concept of non-homogeneous self-assembly might be useful and applicable to self-reconfiguration in UAVs as well; which could be a break through in aerial refueling, and aerial deployment and recovery tasks. The concept of self-assembling smart home architectures is among the vistas as well. Nonetheless the self-assembly and self-reconfiguration problem has been a big source of curiosity and zeal to the people in distributed systems research. Recent years have seen enormous work in terms of algorithmic development for programmable and stochastic self-assembly and self-reconfiguration [12], [13], [14]. Especially tthe prospect of inducing self-assembly behavior via reducing the problem to coordination [15], is very promising. It addresses the question long asked; how can we enable the various elements/agents in a system to achieve a global goal using only local interactions. Working along the same lines, we extend the concept but towards a hardware implementation rather than algorithmic development, by building an autonomous robotic platform with novel features, that can demonstrate the potential of self-assembling behavior via strictly local interactions with the neighboring agents. We intend to utilize the rule sets that result from [15], and apply them to our indigenously designed robots. We believe there is a room of improvement in the deduction of these rule sets, but the desire of implementing the system wins more urgent attention, since a lot since needs to be done along this way. Although the mathematical or algorithmic development is much higher than the existing artificial systems that could actually demonstrate the proposed theory of self-assembling behavior, a variety of small scale robots, and distributed architectures have been proposed over the past several years in attempts to demonstrate the various collective behaviors in multi-agent systems. Some focus on low cost and generalized platforms development that could be used for testing distributed algorithms and decentralized behavior in swarms for instance using localized interactions. [16], [17].
17 These interactions may be of physical nature or not, depending on the design. The physical interaction of the robots with one another and the environment can however be responsible for the self-assembly and self-reconfiguration behavior [18], [19]. This is because the concept of self-assembly is more suited and well understood with the physical interaction of the constituent elements (robots/agents in the artificial case) rather than mere (wireless) consensus. That is why the ever growing interest of roboticists in developing small scale muti-agent systems; that could interact, make or break bonds, physically with their neighbors, and communicate in a completely localized way, continues to flourish. For these desired requirements, magnets and any local communication infrastructure could do the job, enabling the agents to connect to one another both physically and communication-wise. However, the difficulty with using permanent magnets for making bonds among the robots, is that you need to come up with a smart way to undo the attraction, which can be momentum driven for instance [20], but it is not as smooth as you would like it to be. The need for smooth interactions arises especially while implementing a completely autonomous, truly decentralized, and stochastic (i.e. reversible) self-assembly, wheere the robots must possess the ability to make as well as break bonds with their neighbors at will and as fluidly as possible. In this situation, the EPMs appear to come to our rescue [21], [22], but if one wishes to develop something of any practical importance out of self-assembly, then it is not the right way to go. The small sized customized EPMs give rise to problems like lack of self-alignment ability, poor bond strength, and venerability to interference by electromagnetic fields both in latching as well as in local communication. The requirement for a separate control circuitry is another thing. Thus we stick to permanent magnets in our approach, but we come up with a novel latching mechanism by combining the high bond strength of permanent magnets with the high torque and splendid precision of ultra-nano servos. With this unique design, we develop a small
18 robot with autonomous latch and de-latch ability capable to make and break bonds with its identical neighbors and communicate via visible light.
1.2.2
Aerial Grasping
Aerial mode of conveyance undoubtedly holds the elite position in all sorts of autonomous transport. It is quite effective and probably the most reliable way to reach the otherwise inaccessible locations. Recent years have seen great enhancements in design, autonomy, and path planning of unmanned aerial vehicles (UAVs) and robots [23], [24]. This enormous advance in the field of aerial robotics has urged roboticists to design and build robots that are capable of aerial manipulation and tasks including aerial grasping [25], [26], [27], [28], collaborative construction using flying robots [29], and aerial perching on unstructured surfaces [30]. Working along the same lines, we develop a passive magnetic aerial pickup and an impulse based drop mechanism by utilizing the high lift to weight capability of magnetic grippers and extend our previous work [31]. We exploit the fact that although the versatility in aerial grasping [25] remains an interesting topic in research, for nearly all practical and industrial applications, the need to grasp ferrous objects in particular remains a key objective. Given most of the aerial grasping mechanisms put high constraints on the maximum payload limit as well as they constrain the aerial maneuvers of the robot itself, our proposed design is based on maximizing the payload capability while keeping the constraints on the aerial maneuvers of the robot; to a minimum. We introduce a simple passive gripper design for ferrous objects with instantaneous gripping capability. In addition the design possesses the ability to grasp and carry multiple ferrous objects at the same time, while keeping track of the object count via local sensor feedback. For dropping, the proposed design employs the concept of impulse of a force. The specific emphasis on grasping ferrous objects is not just for the ease of imple-
19 mentation but is of significant practical importance. Metal (mainly iron, tin, nickel) containers or enclosures and mounts with ferrous attachments are known for their strength and reliability in all modes of transportation. They can be readily manufactured and attached to or used with almost all kinds of objects that are of practical interest and which for instance, need to be aerially transported. Such ferrous enclosures have been in use for long time, ensuring safe transportation of delicate materials and objects that need to be kept out of any electromagnetic interference. The concept is practically valid for the important social applications as well like medical aid and package delivery [32]. In short, the importance of modern aerial grasping robots cannot be more emphasized. From prestigious robotics competitions [33], to the disaster management robots design challenges [34], it is desired for the robots to possess the ability to grasp and deliver objects of different shapes and textures [26]. The first and the obvious motivation for conceptual design of a grasping mechanism is the human hand itself. Many bio-inspired robotic grippers [35], [27] are based on concepts that are fetched from the human hand. However, its control and motion planning is complex because of its enormously high number of degrees of freedom and rigid nature of its joints. Moreover, it is not very scalable; that is, it is not easy to mount a humanoid arm on an aerial robot for instance without causing much influence on its dynamics. It makes the control of the aerial vehicle very challenging itself. Because of these limitations, people also looked into soft and flexible compliant designs [36], [37], which employ opposing contact forces for their grasping action. For heavy payloads and large objects, this scheme does not seem to be very convincing since a strong attractive force is required in order to pick them up. For non-ferrous objects, in general, the proposed techniques like self-sealing suction [25] and pinch-hole grasping [29], [28] mechanism, have been proposed. These seem to perform well for small objects but in case of heavier objects and especially
20 metallic surfaces with curvature, things can go complicated since shiny or slippery surfaces may cause the suction to fail. Similarly, the pinch-hole claw mechanisms cannot be used for heavy metal objects or the objects that does not have a planar surface in general. Furthermore, the Electro-Permanent Magnet (EPM) technology has desirable features for ferrous grasping. But it has its own limitations. Its grasping is not passive and needs an active control circuit, requires sometime for the magnets to be activated while the gripper has to remain aligned flat with the object, which in many cases is very difficult to achieve. Moreover it is heavy for many aerial vehicles that are relatively small and light weight. Hence, we put our focus on the permanent magnetic pick up and an impulse based drop. The analysis reveals the lift provided by our proposed design is as strong as the dry adhesion [38], giving the gripper a high payload capability. However, the design is much simpler and the system complexity is remarkably small. At the same time, simple yet effective drop mechanism ensures that a contact can be broken as desired, to eventually drop the object.
1.3
Thesis Outline
The following chapters of the thesis are arranged as follows. Chapter 2 details the design and implementation of usBot; a programmable self-assembly robot. We emphasize on its novel attach/detach mechanism, while highlighting its salient performance features, based on experiments. Chapter 3 introduces readers to two novel gripper designs for autonomous aerial grasping of ferrous objects. We describe the design, and key features of both the systems and carry out the performance analysis which is backed by experimental results. In chapter 4, we provide an overall summary of the thesis. In addition to the brief hints on our future directions in respective chapters, we provide a detailed view of the future prospects into both projects as well. It is then followed by the references and appendices.
21
Chapter 2 usBot: A Robot for Programmable Self-Assembly
2.1
Motivation
The desire to design and implement intelligent systems capable of demonstrating programmable self-assembly is driven by a combination of the proposed novel algorithms for stochastic self-assembly, and the impressive latching ability of permanent magnets. On the theoretical side, the primary motivation comes from the game theoretic formulation of self-assembly as a coordination problem in [15] which synthesizes rule sets for agents with a collective goal of a target assembly. This formulation puts certain constraints on the agents that restraints them with the possibility of only minimal local interactions. Here the target assembly is represented by a graph, where each vertex represents an identical agent. If there are more number of agents (robots) available than are required in the target assembly, then as many copies of the assembly shall be formed as possible. Fig. 2.1 shows a sample target assembly represented as a graph with five agents. The formulation in [15] presents variants of two algorithms, one serial Singleton, and one parallel Linchpin, for programmable self-assembly. The former asymptotically, whereas the latter always achieves the maximum yield of the target assembly. Fig. 2.2 shows a quick overview of the algorithms. Each agent maintains its internal state which is represented as a label on the graph. At each time instant, two agents are randomly selected. If a rule from the finite rule set applies to the
22
Figure 2.1: Representation of a target assembly as a graph; the vertices represent agents and the edges define the neighboring relationship in the graph. [15]
Figure 2.2: Overview of the game theoretic approach for programmable self-assembly; blue arrows highlight a forward action (bond formation) while red arrows represent a reverse action (a bond break) or avoidance.
agents, they can either update their states accordingly (hence update the graph) or maintain the old states depending on the probabilities of the forward and backward actions. If there is more than one rule that could be applied to the agents, then one is picked at random. This approach is innovative in the way that it allows probabilistic performance guarantees for the distributed self-assembly in terms of stochastic stability. Another key aspect is its operation under reversibility constraints, which are a necessity in several applications. Since, we do not want the agents to make permanent bonds hence, the requirement to dissemble smoothly is as vital as the assembly itself. Given this motivation, for the thesis, it was thus, of extreme importance to under-
23
Figure 2.3: Simulation results for Singleton and Linchpin, and their non-reversible derivatives for a maximum yield of three. [15]
stand and analyze the algorithms and to translate their constraints appropriately into hardware implications. Thus over the course of this thesis, we reproduced the results presented in [15] using MATLAB. Fig. 2.3 shows the performance of the two algorithms and their non-reversible variants in a test case where the maximum number of target assemblies is three. At the same time, on the hardware end, we propose a smart way of controlling permanent magnets in such a way, that one can achieve their high bond strength as well as use the same to break off the bond with as much repulsive strength, when desired. Thus combining the concept from theory with our innovative design, we come up with usBot (Fig. 2.4), a passive magnetic latching cube for programmable self-assembly. The following section details its design and key features.
24
Figure 2.4: usBot: A passive magnetic latching robot for programmable self-assembly.
2.2
Construction
In this chapter, we introduce usBot; a clean 50 mm cube shaped robot with no external elongations or tethers, capable of demonstrating self-assembly in 2D. At present the aim is not to worry about the 3D assembling capabilities or the self-actuated locomotion of the robot, rather we put our attention on designing intelligent ways to incorporate passive magnetic latching for the bonding and de-bonding behavior. The reason for choosing a closed cubic geometry is for convenience of symmetry and for the ease of self-alignment during the self-assembly. Also, it ensures the homogeneity all around the assemblies, and most importantly this design will enable us to extend the concept to 3D self-assembly, using the pivoting cube model as in [20]. This is a very important feature since probably it is the only modular robot platform [39] that has successfully demonstrated a potential for self-assembly in 3D. So, in order to design an optimum robot with the desired capabilities, as well as the ability to adapt
25 to future prospects, we narrow down the necessary requirements as derived from the theoretical grounds.
2.2.1
Design Constraints
The mandatory design constraints for usBot were enlisted as: Small size centimeter range robot design for future scalability. On-board and autonomous embedded control. Ability to create high strength bonds with other robots. Bond creation should be self-assisted i.e. it should help pull the robots close
together. (preferred) Ability to break high strength bonds with other robots. Bond breaking should be self-assisted i.e. it should help push the robots off.
(preferred) Ability to refuse bonds with other robots i.e. avoidance. (preferred) Reliable neighbor to neighbor communication, when in contact.
In addition to these mandatory constraints, some optional constraints were also set for future work, that are described later in the chapter. The following section presents the design and specifications of the first version of usBot which has been developed completely at RISC Lab. in record time of less than a year which also meets all the above mentioned requirements.
2.2.2
Design & Features
Fig. 2.5 shows both the top side (left) and bottom lateral (right) view of usBot. It is a perfect cube of 50 mm a side length, and consists of two stationary (top and bottom)
26
Figure 2.5: usBot:top side view (left), and bottom lateral view (right) of the design. The circular enclosures at the bottom surface are permanent magnets that assist the bot for its sliding locomotion.
and four active rotating (side) faces. This first version of the robot is constructed with a total of 110 individual parts and electronics components assembled together. All the structural parts are 3D printed separately using the Objet30 Prime . Two printing materials used (for the sake of adding color to the bots) are VeroGrey® and VeroRed® . Both are strong and rigid plastics. As shown in Fig. 2.5, the bottom of the cube is fitted with circular permanent magnets of two different sizes. These help the bot in its sliding locomotion as well as are of key significance for its future self-actuation concept. Fig. 2.6 shows an expanded lateral view of usBot highlighting its various parts. The structure comprises of two symmetric half-frames that can slide into one another and are connected via magnetic supports on both sides. This structure is responsible for holding all the components inside a 50 mm cube and gives the robot its firm shape. The robot is powered by two 3.7V 260mAh LiPo batteries that are connected in parallel. They can operate usBot for about two hours in current de-
27
Figure 2.6: usBot: the expanded lateral view of the design; different parts are labeled numerically; (1) top cover, (2) micro-controller, (3) servo mount, (4) latching magnets, (5) LEDs, (6) servo to rotor mount, (7) face alignment magnets, (8) top half frame, (9) servo, (10) side closure magnets, (11) ambient light sensor, (12) bottom half frame, (13) servo shaft, (14) latching rotor, and (15) battery (hidden view).
28 sign configuration. The batteries along with necessary electronics like regulators and connection interfaces are placed on the base. It is important to note that since it is the first prototype of usBot, we intend to build it using off the shelf components than going fully customized in the first go. It would give us a better idea about the performance of the robot before going for a final finished product. Each side face of the cube is fitted with eight symmetrically arranged bipolar 2 mm x 1 mm discs of neodymium magnets. These help align the bots together and ensure smooth latching. Two pairs of glsplled and ambient light sensors are mounted in opposite positions in a symmetric fashion over the corners of the each active face of the robot. This pattern ensures that the two facing robots will always have two aligned pairs of LEDs and light sensors established during contact. In fact we intend to use it as a Tx/Rx channel for local communication among the robots. It is in fact a very efficient way of local communication among the neighboring robots. Apart from sensors, batteries, and necessary electronics, each of the two half frames supports a servo mount which in turn keeps two servos in place each at right angle to the other. The servos are in turn connected to the latching rotor via a rotor mount in such a way that the whole servo assembly can rotate with its center coinciding with the center of the corresponding face. Each rotor encloses four latching magnets on the inner side.These are strong 10 mm x 1 mm discs of neodymium magnets, each capable of having a pull of 0.51 kg. These magnets and the servo-latching rotor mechanism is the basis for our self-assembling robot design. The principle and working of the latching mechanism is explained in the following section. The brain of the robot, the microcontroller is located at the top. It is an Atmega328 based board with 32 KB of flash memory. In addition it comes with a variety of serial, and digital and analog I/O interfaces, and useful debugging capabilities. Its 16 MHz, 8 bit Atmel processor and the memory resources are sufficient for our design requirements. It can be programmed via its built-in bootloader using an ordinary FTDI cable. We
29
S. No. 1. 2. 3. 4. 5. 6. 7. 7. 8. 9. 10. 11. 12.
Table 2.1: usBot: List of Components Component Quantity Atmega 328 (Arduino pro mini 16MHz/5V) 1 Ambient light sensor TEMT6000 8 SMD blue LEDs 8 5V Step-Up Voltage Regulator U3V12F5 1 Hitech ultra nano servo HS-35HD 4 Nanotech 1S, 260mAh LiPo battery 2 Low battery voltage sensor 1 10 mm x 1 mm Neodymium disc magnets 16 5 mm x 1 mm Neodymium disc magnets 10 25 mm x 1 mm Neodymium disc magnets 1 2 mm x 1 mm Neodymium disc magnets 44 Miscellaneous wires & connectors N/A 3D printed structural parts N/A
plan to develop a wireless communication framework for the bots in future so that the program can be downloaded wirelessly as well as the robots can send individual updates on their internal states to a base station. Table 2.1 provides a list of all the components and their respective quantity used in a single robot. The completely assembled usBot currently weighs 95 g. For now, we fabricated a total of six robots and the per unit cost of usBot is around 200 USD. We expect it to go down to around 150 USD in a mass production of say 100 bots.
2.2.3
The Latching Mechanism
This is the most important part of the usBot. The novelty of our design lies in its simplicity yet high efficiency. The principle behind the working of the latching mechanism is derived from classical Physics and the laws of magnetism. If a ferrous object is in close vicinity (from a few mm to few cm depending on the object) to a permanent magnet, there exists a force of attraction between the object and the magnet. Mathematically the force of attraction of a magnet at its air gap (the space
30
Figure 2.7: usBot: the enlarged (a) front/outside, (b) back/inside, and (c) back (side of the latching rotor) views of the latching mechanism; different parts are labeled numerically; (1) servo mount, (2) latching magnets, (3) mechanical brake for overrotation, (4) servo, (5) servo to rotor mount, (6) servo shaft, and (7) mounting hole for servo mount.
around the poles of a magnet) is given by Maxwell equation:
F =
B2A 2µo
(2.1)
where F is the force (N), A is the surface area of the pole of the magnet (m2 ), B is the magnetic flux density (T) and µo is the permeability of air. Thus if the target is a magnet itself, then there exists either a force of attraction or repulsion between them. The nature of this force depends on the polarity of the two approaching magnets. Nevertheless, this force is almost twice (in case of neodymium magnets) as compared to the magnetic force given by Eq. 2.1. We employ this concept in our quest to achieve programmable self-assembly. In the latching rotor the magnetic polarities of adjacent magnets along the circumference are always kept different. In this way a complete reversal of all magnets polarity can be achieved by mere 90 degrees of rotation, either clockwise or counter-clockwise. This
31 sounds a very simple concept and in fact it is that way, but its extremely effective. Refer to Fig. 2.7(c), which shows the initial or default position of the latching rotor from behind. At this configuration, if two robots approach each other they are going to attracted towards each other with a force roughly eight times the pull of a single magnet. This bond formed among the robot faces is strong and yet not permanent, in the way that it can be easily undone using a 90 degrees rotation by either of the bots. The only issue is that this action of bond breaking by revolving either of the latch requires high torque1 . Due to small size constraints on the robot design, and difficulty of finding small size and high torque servos, we introduce a 1 mm shielding on either sides of the bond. This plastic decreases the magnetic force of attraction to about 8 N. At this level the bond can be broken by the 90 degrees rotation of the smallest high torque servo2 . It is interesting to note here that this latching mechanism is completely self-assisting i.e. it can pull the bots close as well as push them away depending on the rotor rotation. To ensure reliability and consistency in bonding/de-bonding action, the latching rotor has two mechanical braking arms along its diameter to avoid any over rotation that might be caused by a servo slip for instance.
2.3
External Actuation
Since the current version of usBots does not possess the capability of self-actuation of locomotion, we designed a simple external actuation mechanism for the interaction of robots and to perform basic tests and check and validate the primary capabilities of the bots. Fig. 2.8 shows the external actuation platform for usBots. It is a very simple 2D gimbaled platform that has a square 1.5 ft a side, smooth top, which has the ability to move along the X, Y or both axes simultaneously from angles ranging from -45 to 45 degrees. Basically it is more like a random mixing platform for the 1
We designed the bond breaking torque to be equal to 0.065 Nm which is great for such a small robot but still less than the maximum torque of the servo we used. 2 Hitech Ultra Nano servo HS-35HD. It has a maximum torque of 0.0785 Nm.
32
Figure 2.8: The external actuation platform for usBot locomotion.
very first static version of usBots. The setup comprises of two high torque, high speed servos that are mounted at right angles to each other and are supported by the base. Each of the two servos is independently controllable, and is calibrated to move in between the designated angles range. Rest of the components include a battery, a voltage regulator, and a microcontroller to control the servos. The goal is to allow the usBots placed on the top to mix up and interact with each other in a random fashion, so that they can assemble/dissemble depending on the algorithms under consideration. A proportional controller controls the movement of the platform in the xy-plane and ensures a smooth change in the orientation over time. We found out during the tests that for temporary purposes, this simple setup works perfectly fine and serves the purpose really well.
33
Figure 2.9: The force test results for usBots; blue: face to face attraction, red : face slide attraction, and green: face to face repulsion.
2.4
Experiments
As previously discussed in the introduction chapter, the latching mechanism and the face magnets need to provide enough force for robust bonding between two robots. Similarly, the servo needs to produce enough torque to successfully break the bond. Thus we performed a series of tests in order to analyze the performance of our design. It was observed during the experiments that the torque required to break the bond is close to the maximum torque limit of the servo. This ensures a strong repulsion at bond breakage which not only breaks the bond but also pushes the robots away. We also analyzed the performance of the design in repelling other bots in an unwanted state by avoiding the bond. This is a characteristic which is unique to our design and hence it is of significant importance. During the tests we found out that despite the face alignment magnets, that are attractive in effect, the latching mechanism is quite effective in keeping the other robot off the face by the repulsion of latching magnets. Fig. 2.9 and Fig. 2.10 showcase the results for force test and torque test respectively.
34
Figure 2.10: Relationship between the bond state and the applied torque; State 0 means the bond holds while state 1 means it is broken. The blue trace represents the response of the actual latching mechanism to applied torque. the red trace represents the maximum allowed torque of the servo used.
We did some tests to check the reliability of local communication based on our LED/ambient light sensor pair. Since the data that is required to be communicated is very small, there could be multiple ways of doing it. The light sensor has an analog output which can however be used in digital fashion as well. The reason for having two pairs of LEDs and light sensors is redundancy; i.e. one of the two pairs is used for confirmation of bond formation or breaking. It also helps to make the communication between the two bots faces synchronized. Nevertheless, each face’s LED (Tx)/light sensor (Rx) pair is simply connected to the controller via I2 C protocol. It works perfectly fine and we found out the success rate of communication as 100%. In the future we might be interested in some better and more sophisticated sensors and communication protocol to boost its speed in case of large number of agents, especially if we migrate to 3D self-assembly. As far as the self-assembly experiments are concerned, out of the six robots manufactured, we have only two of them fully assembled for now, thus we could only
35
Figure 2.11: usBots: screenshots from the experiments showing the primary moves for self-assembly behavior; (1) bond formation, (2) bond break, and (3) avoidance.
test basic algorithms and communication on our usBot platform at the moment. We did some experiments verifying the ability of usBots to successfully bond, de-bond, and also avoid bonds with the other bot (Fig. 2.11). The results for the conducted experiments were very encouraging and we look forward to assemble all the fabricated bots so that a full implementation of Linchpin [15] can be observed. Fig. 2.12 shows the work in progress while assembling the printed usBots.
Figure 2.12: usBots: work in progress, while assembling rest of the robots.
36
Chapter 3 Passive Aerial Grasping of Ferrous Objects
3.1
Motivation
Aerial grasping is one of the many interesting research problems in the class of aerial robotics. It would be a notable accomplishment if we could successfully combine the ability of drones to access the otherwise inaccessible areas using UGVs, with aerial grasping, perching, and similar aerial manipulation tasks. It would make the dream of carrying out missions like aerial deployment and recovery, and search and rescue in fully autonomous fashion, a reality. Our specific emphasis here, on aerial grasping of ferrous objects is not just for the ease of implementation but is of significant practical importance. Metal (mainly iron, tin, nickel) containers or enclosures and mounts with ferrous attachments are known for their strength and reliability in all modes of transportation. They can be readily manufactured and attached to or used with almost all kinds of objects that are of any practical interest and which for instance, need to be aerially transported. Such ferrous enclosures have been in use for long time, ensuring safe transportation of delicate materials and sensitive objects that need to be kept out of any electromagnetic interference. Hence we emphasize on the design of passive magnetic grippers for UAVs. This chapter details two designs of the passive magnetic grippers fabricated at RISC Laboratory for indoor and outdoor aerial grasping using the quadrotors.
37
3.2
Working Principle
Simple Physics laws are the basis for our proposed gripper. The novelty of our design lies in its simplicity yet high efficiency. Picking objects up is straightforward; if a ferrous object is in close vicinity (from a few mm to few cm depending on the object) to a permanent magnet, there exists a force of attraction between the object and the magnet. Mathematically the force of attraction of a magnet at its air gap (the space around the poles of a magnet) is given by Maxwell equation:
F =
B2A 2µo
(3.1)
where F is the force (N), A is the surface area of the pole of the magnet (m2 ), B is the magnetic flux density (T) and µo is the permeability of air. Thus if a magnet acts vertically, the mass m it can lift successfully in (kg) is given as: m=
B2A 2µo g
(3.2)
where g is the gravitational acceleration (m/s2 ). As far as our gripper is concerned, no external activation is required for picking things up, since the magnets are always activated being permanent (passive) in nature. So, the grasping action is spontaneous. The dropping mechanism however is of more interest. It is based on the concept of impulse of a force. In fact, it is the result of a direct implication of Newton’s second law of motion: Favg = maavg = m
δv δt
(3.3)
where Favg is the average force acting on the object (N), m is the mass of the object (kg), aavg is the average acceleration (m/s2 ), δv is the change in its velocity (m/s) and δt is the time of action of the force (s).
38 Impulse of a force J (Ns) is defined as the product of the average force Favg and its time of action δt. It can therefore be represented as a change in linear momentum of the object, to which the force is applied. Mathematically:
J = Favg δt = mδv
(3.4)
Hence, for instance, to achieve a desired increase in momentum of the object, one can either increase the force or decrease the time of action and vice versa. This concept is the key in our design for dropping mechanism. Using an actuator and combining it with the natural pull of gravitation which is always acting downwards, we utilize a sudden push to undo the attraction of the magnets that is gripping the object, and to force throw the object down. This scheme sounds so simple which is the fact, but it is surprisingly effective. In practice, as we found out later in our experiments that we do not even need a very powerful actuation force, thanks to the weight of the object itself. Thus, in this scenario, the heavier the object is, the easier it is to drop it using a smaller force and vice versa.
3.3 3.3.1
Passive Magnetic Gripper Gripper Design
Fig. 3.1 shows an expanded lateral view of the gripper. Each of the various parts shown is 3D printed separately using the Objet30 Prime . The material used for printing is VeroClear® which is a clear plastic. In order to distinguish different parts, they were later on sprayed with acrylic paint. Its lower end spans over a circle having a diameter of about 100 mm. Three magnet enclosure pads are mounted to the main frame on the lower side. Each of the three pads is in turn fitted with three permanent magnets. These are 6.33 mm ( 14 in) cubes of neodymium with each capable of providing a pull of approximately 10 N (1 kg). Each cube itself weighs only 1.90
39
Figure 3.1: Passive gripper design for ferrous objects; various parts are labeled numerically. (1) gripper mount, (2) flexible rubber connector based on compliance, (3) servo plate, (4) camera mount, (5) camera, (6) magnet enclosure pad, (7) digital infrared sensor, (8) magnets, (9) drop plate, (10) main frame, (11) servo, and (12) servo shaft.
40 g. Hence, for aerial applications, where payload is a matter of high consideration, these magnets are of vital importance in our gripper design because of their high pull to weight ratio. For our design, we intended to develop a gripper capable of grasping objects ranging from few tens of grams up to one kilogram. Hence, we used nine of such magnets in a symmetric triangular arrangement of three each. In theory, it enables the gripper to support a payload of about 9 kg. In practice however, it is much less than that for many practical reasons. First the object to be carried is not necessarily carried by all the magnets together. Secondly because of the payload capabilities of the aerial vehicle, effects of gravity, aerial disturbances and the dynamics of the aerial vehicle, it is not guaranteed that all magnets pull the object with the same force. Hence the net pull of the gripper is not equal to the sum of individual pulls of all the magnets in general. The gripping surface is a symmetric triangle with each individual magnetic grip separated by a distance of 75 mm. This separation makes it capable of grasping multiple objects as well. The drop plate is designed in a way that ensures it is in good contact with an object all around the gripping pads, to push it off when activated. As indicated in Fig. 3.1 each of the three inner corners of the drop plate is equipped with a digital infrared sensor. It is thus able to provide discrete information about whether the object has been successfully grasped or dropped. Whereas in case of multiple objects this feedback information can be used to keep track of the object count and hence the free location on the gripper can be decided to pick new objects accordingly. This attribute is vital for autonomous aerial grasping in the absence of any human feedback, which is the ultimate goal of all sorts of aerial transportation. The drop plate is further connected to the camera hold-servo connect via three triangular symmetric rods. These are designed to slide against the inside of the main frame and can thus move freely up and down as controlled by the servo. During the pick up operation the servo stays idle and the drop plate thus stays above
41 the gripping pads, allowing the objects to be grasped by the magnets. When required to drop the objects, the servo is activated and it pushes the drop plate down via its shaft connected to the camera plate. The action time for this push is less than a tenth of a second. Hence, the impulse generated by the servo is sufficient to push the object off the gripper. The servo is kept in place by a plate at the top which also holds the main frame. Because of the permanent nature of the magnets, there is no need to accommodate any angular misalignment at the local level at individual gripper pads. A 3D compliance is however employed at the top, using three layers of 5 mm silicone rubber fused together with a central plastic rod that limits its pitch and roll deviations. Basically it serves as a flexible connector between the gripper and the mount and allows the gripper to swing up and down by as much as 20 degrees, enabling it to cope with any angular misalignment at the global level. The camera plate also has an attached camera centered inside the drop plate. The design enables the camera view to stay clear as long as the object is not grabbed.
3.3.2
Gripper Assembly
Due to several moving parts and embedded electronic components, appropriate assembly of the gripper is vital for its optimum performance. Hence, with 3D printing done and support material removed, first of all, magnets were glued inside the pads. It was ensured that all the three magnets are well aligned and leveled at the same surface. The pads were then glued to the main frame with an additional support from two 1.5 mm plastic screws. The sensors were glued to the drop plate at the corners after testing them individually. The drop plate was then slid inside the main frame from below. This was the most important part of the assembly since we needed this motion to be as smooth as possible to put no extra strain on the servo. Since 3D printing requires removal of the support material and cleaning afterwards, special attention was paid to the cleaning of these two parts. We also applied some silicone
42
Figure 3.2: The gripper assembled and mounted on the quadrotor frame for testing; (1) drop plate in normal position, (2) drop plate in activated position for dropping.
based lubricant in between, in order to make this movement as frictionless as possible. The camera was attached to its plate with four 1.5 mm screws, which was then glued as well as screwed onto the drop plate. This sub-assembly could thus freely slide up and down along the main frame. Then we glued the servo in place and connected its shaft to the drop plate via the camera-servo connector. Finally the whole gripper was attached to its mount via a central M3 screw though the rubber central compliance. Two more M3 screws attach the whole gripper assembly to the quadrotor. Fig. 3.2 shows the actual gripper assembled and mounted on the quadrotor frame.
3.3.3
Quadrotor and Control System
The gripper was mounted on a Q450 V3 quadrotor frame from HobbyKing® which along with the landing gear, the battery and all on-board electronics weighs 1236 g. After including the gripper, the camera and its mount the total mass becomes 1582 g. This leaves the quadrotor with an additional payload capability of about 600 g. Fig. 3.3 shows the relative contribution of various components in total mass of the aerial vehicle. The gripper was mounted at the center of the quadrotor as closely as possible to keep its dynamics in tact. The camera was connected to Odroid via USB
43
Figure 3.3: Relative breakdown of mass of the aerial vehicle. while the sensors and the servo were plugged into Pixhawk autopilot system. Odroid sends high level algorithmic commands using ROS and C++/Python based firmware, to Pixhawk which then directs the quadrotor to perform the desired maneuver by varying the thrust signals to the four rotors. For its autonomous operation, the system relies on some kind of localization or coordinate system to determine its position and velocity in space. It can either be a Global Positioning System (GPS), differential GPS, or an indoors motion capture system. For the sake of initial testing and experiments, we used the indoor motion capture system for position and velocity control in the RISC laboratory. Markers were placed on the quadrotor as well as the objects and the drop zone to specify their coordinates. This enabled us to grab the objects from their respective locations and drop them at the designated drop zone. Fig. 3.4 shows the quadrotor with the gripper mounted along with all the on-board electronic components.
44
Figure 3.4: Top side view of the quadrotor assembled with the gripper and all onboard electronics for testing.
3.3.4
Experiments
Maximum Payload Two different experiments were performed in order to estimate the maximum payload limit for the gripper. First the gripper was mounted vertically on a tripod stand and a ferrous plate of known mass was attached to the magnetic grip. We let the plate be gripped by all the three gripping pads simultaneously. A sensitive spring balance of known mass was then connected to the plate. The spring balance was then continuously pulled down by applying an incremental force. The force was applied downwards at a right angle to the gripper surface by means of a continuous rotation high torque servo via an inextensible string. We were interested in finding the maximum pull that breaks the contact and makes the plate fall off. This force when added to the mass of the plate and spring balance itself gives the maximum payload for the gripper. We repeated the experiment ten times and found the mean maximum payload to be 12.5 N (1.3 kg). This is quite impressive given the gripper
45
Figure 3.5: Relationship between gripper state and the applied force (payload) for the gripper; (1) all three pads, (2) a single pad only. State ’0’ means the payload is attached while state ’1’ means it is dropped. The hysteresis region indicates the variation in various observations. Blue trace represents the min while red represents the max.
itself weighs only 75 g. In the second experiment, we went through the same procedure for each of the three individual gripping pads. The purpose was to investigate the individual payload capability of a single pad alone, in order to analyze the performance of the gripper in grasping multiple objects. Again, we repeated the experiment ten times for each pad and found out the mean maximum payload to be 4.5 N (0.5 kg). Fig. 3.5 shows the results for both the experiments.
46
Figure 3.6: Percentage of successful drops in relation to the payload. The success rate sets a lower limit of about 50 g to the payload range for our current gripper design.
Minimum Drop Threshold We also tested the ability of our gripper to successfully drop objects based on their weight. We found out that the dropping mechanism operates perfectly for the objects ranging from average up to the maximum payload limit for the gripper. However, we were interested in analyzing whether there exists a lower limit of allowed payloads as well. So, we reduced the payload on the gripper in steps of 10 g from the maximum payload, all the way down to 10 g. The results were quite interesting. The gripper was able to successfully drop all the objects from 500 g down to 70 g with a success rate of 100%. However, below 50 g, we faced some significant trouble dropping objects. This was because the servo and the weight of the object combined were not enough to overcome the attractive force of the magnets. The problem however can be easily avoided using a stronger servo. Fig. 3.6 shows the relative performance of the drop mechanism for a given range of payload.
Maximum Sustainable Slide Sudden obstacle avoidance or attitude change can be tricky while the quadrotor is carrying some payload. In such situations, the object can be thrown away due to the
47
Figure 3.7: Experimental setup for determining the worst case maximum sustainable slide for the maximum, average and minimum payload.
sliding effect of Coriolis force. Hence, in order to find out the limits of our design, we performed lateral force tests. These consisted of measuring the force up to which the object will remain attached to the gripper without sliding off, and surprisingly the results were quite encouraging for such a small grasping mechanism. It is important to note here that greater the mass of the object, the greater is the Coriolis (slide) force it will experience when the quadrotor performs an aggressive roll or pitch. We used the data about the payload range from previous experiments and investigated the possible slide for the three test payloads one by one. Same experimental setup was used as in the last experiments with the only difference that we performed the experiments in a position as if the quadrotor were performing a 90 degrees roll or pitch. We picked the minimum force at this instant as the threshold value for the maximum sustainable slide. It is the worst case scenario in the sense that the weight of the object itself gets aligned with the Coriolis force which tends to slide it down. Fig. 3.8 shows the experimental results for the three test loads. The maximum load (500 g) did not require any additional force. That is its weight was enough to make
48
Figure 3.8: Maximum worst case sustainable slide test results for the gripper; red is for the max payload (500 g) showing worst case sustainable slide is zero; green is for the average payload (275 g) indicating a worst case slide force of 3.6 N while blue represents the worst case max sustainable slide of 5.9 N for the minimum load (50 g). The smaller the mass of the object, less is the effect of Coriolis force and hence higher is the maximum slide, it can sustain.
it slide down at the worst case position. It shows that with a payload of this order one can not perform a flip over using the quadrotor while expecting the payload to stay intact with this gripper design. However, it was observed during the experiments that the rubber compliance connection at the top of gripper was not only important for handling the misalignment issues but was also vital for preventing the payload from sliding. Because of this flexible mounting to the frame, it never lets the weight of the object to align with the Coriolis force which is acting along the direction of slide. Hence, the actual slide force is always less in our design than it would have been if the gripper mount was rigid. It thus provides a safety margin to our gripper design in terms of payload slide avoidance.
Aerial Grasping Finally we tested our gripper to grasp and drop different ferrous objects. Specifically, we wanted to demonstrate its ability to grasp a single object using one, two
49
Figure 3.9: The quadrotor grasping: (1) metallic box, (2) scissors, (3) screw driver, (4) scissors and screw driver, (5) three metallic boxes together; each on one pad, (6) spray paint can vertically, (7) spray paint can along the curved edge using two pads, (8) two spray cans together, and (9) small quadrotor (FPV-180).
or all three gripping pads, and multiple objects using two and three magnetic pads along with the picking up and dropping of uneven, inclined or curved ferrous objects. All these tests were performed inside the RISC laboratory arena using the indoors motion capture system. Each of the objects was placed in the marked location. For ease of mobility, the objects were placed at least 30 cm apart from one another. A drop zone was marked at a corner in the arena. The quadrotor was set to fly at an altitude of 1 m. When commanded, the quadrotor would fly to the location of a specific object and descend down to grab it, and fly back up to 1 m. Then it would either go pick up another object or fly to the drop zone to drop it. Fig. 3.9 shows the snapshots of the quadrotor while grasping various test objects. Small and light weight objects needed only one pad. The cubic box required all the three pads to grab it at its center. We were also able to demonstrate that it was possible to pick one box first and then grab another later using another pad. Then
50 we tried with some heavier and curved objects like an acrylic spray can. First we grabbed it vertically while it used all the three pads on the gripper. Later we tried to grab it from the curved side. Again, the gripper was able to successfully grasp the can using two of the three pads. Further we also attempted to grab a second can while the gripper was already carrying one. The gripper was able to grip and fly up with both the cans attached. Finally we used the gripper to pick up and drop a small quadrotor (FPV-180) from one place to another. For all the test objects used, the dropping was successfully achieved and the drop mechanism worked fine. Also, while carrying the objects we occasionally performed some aggressive maneuvers with the quadrotor involving sudden change of attitude using manual flight mode. None of the objects flew off or dropped because of the slide effect discussed earlier. One observation while grasping multiple objects was, the objects should be picked in the order of their increasing heights to avoid any collision with already picked ones.
3.4
Heavy Duty Gripper
The earlier design [31] performed significantly well in terms of indoor tests with a small quadrotor, but for some practical reasons and its mechanical limitations, we did not find it well suited for autonomous grasping applications outdoors. First being the camera has to be stabilized using a 2D gimbal all the time in order to track, pick up, and drop the target objects. Further, the environment is much stable indoors, and with a high precision motion capture positioning, there is no hard constraint on the camera to be stabilized all the time for semi-autonomous operation of the quadrotor. However with reduced accuracy of Global Positioning System (GPS) outdoors, and high reliability on vision for fully autonomous operation, aerial grasping gets tricky. Wind is another aspect that can cause trouble in the field. Previous design did not leave enough space for the gimbal mechanism to be mounted for the camera. Also,
51 mounting the whole gripper on a gimbal would not be feasible since a small gimbal may be able to stabilize the camera with no payload but at full payload, it is very likely to stall. Another limitation was found in the pick up feedback mechanism. The infrared sensors worked fine indoors but under different lighting conditions outdoors they were not reliable. Moreover we wanted to make its payload range wide. The goal was to reduce the minimum drop threshold payload value to zero and increase the maximum payload limit while keeping the weight of the gripper as low as possible. All these factors combined, they served as a motivation for us to design a more robust, stable, and heavy duty gripper for autonomous aerial grasping.
3.4.1
New Gripper Design
Fig. 3.10 shows both (a) top side, and (b) bottom side lateral views of the gripper. Each individual part is 3D printed using the Objet30 Prime from Stratasys® . The gripper base spans over a circle of diameter 200 mm. Four magnetic enclosure pads form the basis of the gripper’s grasping ability and are mounted at the bottom, 180 mm apart, and at right angles to one another. The gripper base looks like a symmetric cross. All the parts shown in the gripper assembly, other than the magnetic pads are printed using the VeroWhite® material, which is a rigid plastic. However, for the four magnetic pads TangoWhite® material is used, which is a rubber like plastic. This is to ensure that the contact between the gripper and the objects is damped and the grasping action is smooth. It is discussed later in the chapter that these flexible pads also increase the maximum sustainable slide force for the gripper. Unlike previous design [31], each of the four magnetic pads is further fitted with four, instead of three, 6.33 mm ( 41 in) neodymium cubic magnets, at the corners, and a digital push button at the center. Also, the magnets are mounted with some separation distance between each other on every enclosure pad. We observed previously that making the magnets touch each other and gluing them together reduces the net
52
Figure 3.10: Heavy duty gripper design for passive aerial grasping of ferrous objects; (a) top side view, (b) bottom side view; various parts are labeled numerically: (1) cable passage for camera, (2) mount plate, (3) separable support, (4) holding plate, (5) connecting spacer, (6) drop plate, (7) camera plate, (8) servo plate, (9) servo mount, (10) magnet, (11) drop plate support, (12) camera fitting, (13) push button, and (14) flexible magnetic pad.
53 lift of the magnets combined, compared to the sum of their individual lifts. This is because of the misalignment/ cancellation of magnetic fields of each magnet by surrounding magnets field. So, in this new design, we keep them separate to utilize as much as possible, the capabilities of the magnets. The push buttons provide a reliable pick up and drop feedback irrespective of the lighting conditions or the environment. Also, in this gripper we embed the pick up/drop feedback into the pad instead of the drop plate, because in case of multiple small objects grasping, it is much more reliable, and ensures flawless autonomy with no erroneous feedback message to the controller, which could occasionally result before. With this configuration, the gripper is capable to pick up to four objects at the same time, using all of its four magnetic pads. The push button feedback allows the controller to keep track of the status of each pad, and the position of vacant and used magnetic pads in case of multiple objects, so it can decide to pick up new objects using the free pads. The drop plate design ensures that it is all around the magnetic pads, and the attached objects; so that it is able to push the attached object off the gripper, when activated. The drop plate has a built in camera1 housing that can be used for vision based tracking and autonomous grasping of objects. The purpose of placing the camera right at the center of the gripper base is to ensure a clear view from the camera all the time, and to help align and center over the target object. The pads are mounted on to the holding plate via 5 mm circular aluminum spacers using 2.5 mm screws. The holding plate has two purposes; one to connect the pads to the servo plate via four symmetric separable support, and second to keep the drop plate and the camera plate in place while allowing them to slide up and down controlled by the servo2 . The separable supports are mounted to the servo plate on top and to the holding plate on bottom via two 5 mm circular screw connects of aluminum. The 1
OV5640 full HD mini 5MP AF USB Camera Module; color CMOS sensor. We used two high speed, high torque servos from HiTEC® ; HS-MS7990TH. It has a maximum torque of 44 kgcm (4.31 Nm), and a speed of 0.21 sec/60deg 2
54
Figure 3.11: Gripper drop mechanism; drop plate (a) in normal position, and (b) in activated (drop) position.
Figure 3.12: Gripper drop mechanism; drop plate (a) in normal position, and (b) in activated (drop) position.
hollow design of the separable supports, its thick 6.5 mm walls with two 45 degrees turns, and aluminum connectors ensure strength and durability of the gripper, and make it capable of withstanding an impact up to several tens of Newtons (N) in case of aggressive grasping or touch down in case of accident. The camera plate is connected via a connecting shaft to the servos which are mounted to the servo plate via four standard 2.5 mm screws. The servos stay in idle mode during the picking operation with the drop plate staying above the pads, thus allowing the pads to grasp ferrous objects. When the servos are activated, they force the drop plate down via its shaft connected to the camera plate. The action time for this push is only 0.16 seconds for 45 degrees of servo rotation. Because of this small time of action, the impulse generated by this actuation is enough to push the attached object off the gripper pads. Fig. 3.11 shows a close up view of the design while Fig. 3.12 of the drop plate in normal as well as activated positions.
55
Figure 3.13: The heavy duty passive magnetic gripper assembled and equipped with all the electronics.
Similar to previous design, due to the permanent nature of the magnets, there is no need to accommodate any angular misalignment at the local level at individual gripper pads. Since the magnets are always active, the gripping action is spontaneous. However, still the flexible gripping pads can take care of small misalignments up to 10 degrees. Unlike our previous design [31], we did not use any 3D compliance on the global level for this gripper. This is explained in detail later in this chapter, since we used a customized heavy duty gimbal to mount the gripper to the quadrotor. This also explains the reason why we made the embedded camera fixed rather than gimbaled in the first place.
3.4.2
Gripper Assembly
Adequate handling of the gripper parts after their printing is of significant importance to ensure proper working of the embedded electronics. Further, to ensure smooth movement of the drop plate, special care is required in assembling the gripper. So after post-printing clean up , first of all, magnetic pads were assembled. The alignment of the push buttons with the magnets is very important. It requires special attention, since one need to ensure, the buttons are pressed when an object is grasped by the
56 pad and released otherwise. Also the buttons should not be too below the surface of contact so that they do not affect the magnetic strength of pads. The push buttons were individually tested for the feedback as well. Also the alignment of the four magnets was ensured before gluing them in their respective places. The pads were then mounted to the holding plate via 5 mm aluminum spacers and standard 2.5 mm screws. The four separable supports were then attached on top of the holding plate via spacers and 2.5 mm screws. The servo was mounted to its plate and the shaft was connected to the camera plate via connector. The servo was checked for its normal operation and its ability to move the camera plate. The servo plate along with the servo attached was then screwed to the top edges of the four supports via 2.5 mm screws. Finally the drop plate was mounted with the camera at its center and then placed into position. The camera plate was then connected to the drop plate via four 35 mm long, and 5 mm in diameter polished aluminum rods. These rods pass through the four corresponding holes in the holding plate, in between which serve as cylindrical openings thus letting this drop sub-assembly slide up or down, as controlled by the servo. The whole gripper assembly was then mounted to the quadrotor using the mount plate via four M3 screws. Fig. 3.13 shows the actual gripper ready to be mounted onto the quadrotor. The gripper when completely assembled and ready to mount weighs around 700 g. We show later that this high weight is however justified by its excellent payload capability especially in comparison to the EPM based gripper.
3.4.3
The Quadrotor
The gripper was mounted on an extraordinary quadrotor; specifically designed for outdoor tasks, aerial grasping being one of them. Fig. 3.14 shows the custom made quadrotor, whose frame has been designed and fabricated by FalconViz® at KAUST. It is a big quadrotor with a symmetric cross configuration with a side length of 85 cm (from rotor to rotor). An entirely carbon fiber built frame, and propellers ensure
57
Figure 3.14: The quadrotor fully equipped with the gripper and other components, and ready for testing.
great strength and makes it perfect for flying outdoors. It is powered by a 22000 mAh, 6S LiPo battery (which weighs 2.5 kg) that can assure a flight time of half an hour with full payload that is 3.5 kg. The quadrotor weighs about 5 kg in total with the battery and all the mounted electronics. Thus the proposed heavy duty gripper is only about 5.5% of the total weight of the aerial robot. The system is equipped with state of the art components for its optimum performance. The quadrotor carries a NUC -i7 computer from intel® which does all the computation required for its autonomous operation. It is a linux machine that sends high level algorithmic commands based on ROS and C++ based firmware, to Pixhawk, the autopilot system which then performs the desired maneuver by varying the control inputs to the rotors. To ensure interruption-less and safe power supply, the NUC is powered using a power converter connected directly from the battery. In addition, the quadrotor is equipped with a number of useful modules, sensors, and components that are necessary for its fully autonomous operation; for example GPS, LiDAR, and a Terra-ranger module for altitude control, and most importantly
58
Figure 3.15: The customized gimbal plate mounted via two servos and a PixFalcon, at the base of the quadrotor.
a customized heavy duty 2D gimbal at the base of the quadrotor. It consists of two high speed, high torque, power servos3 that are fixed at right angles in 2D plane to form a gimbal. The gimbal is controlled via a dedicated PixFalcon controller which is calibrated using Mission Planner interface. Fig. 3.15 shows the gimbal plate, the servos, and the PixFalcon mounted on the quadrotor base. This plate can be used to mount not only cameras but also the gripper, since the servos are strong enough to sustain high torque even at maximum payload. This is why we kept the camera in the gripper fixed; this structure design not only helps to keep the camera aligned with the ground all the time but also nearly eliminates the slide (Coriolis) force completely that might act on the payload. This concept is elaborated in sufficient detail in section V. Another point to be noted here is that in fact we mounted an additional camera on the side of the gimbal plate. This is because once the object is picked up, the view from the central gripper camera is blocked and thus in order to ensure autonomous HiTEC® ; HS-MS7990TH. It has a maximum torque of 44 kgcm (4.31 Nm), and a speed of 0.21 sec/60deg 3
59 operation the quadrotor needs to switch to the second camera in order to find its way to the drop zone. Thus the gripper camera is used for centering, alignment and grasping the objects while the second camera is required for the transport and drop routines. The ROS state machine running in NUC takes care of this state switching operation. There is an additional microcontroller4 mounted on the upper side of the gimbal plate. It is dedicated to read and make decisions about whether the objects have been picked up or dropped, and then publish its decision to NUC via a ROS node. Currently, the main state machine running in NUC subscribes to the decision continuously and holds the authority to override the gripper node decision if necessary. The reason for using so many separate controllers on one quadrotor is to distribute the control as much as possible. Its also good from safety point of view since it can help in avoiding major accidents in case of failures of a single component device.
3.4.4
Experiments
In order to verify and validate the expected enhanced performance of our proposed design, and to compare it with the previous design as well as other competing technologies, we performed a series of experiments and tests. These provided us with useful parameters for the gripper such as the maximum payload, the maximum sustainable slide, and the minimum drop threshold. These parameters define a performance index for the design and enable us to set some constraints for its adequate operation. In addition, we also performed autonomous aerial pick up and drop tests with the quadrotor.
Maximum Payload We analyzed the maximum payload capability of the gripper in two different ways; one for each pad individually, and second for the gripper as a whole with all pads used 4
Arduino Pro Mini board; it is powered via a USB cable connected to NUC ’s power USB port.
60 together. For each test we mounted the gripper to a tripod in a vertical position. A test object was then attached to either a single pad or all the pads. The test object was a galvanized iron sheet disc of diameter 20 cm. The test object was then pulled down slowly by a high torque low speed continuous actuator via a sensitive digital spring balance. So the value of the balance at any time plus the weight of the test object equals the effective payload attached to the gripper. We continued to increase the applied force in steps of 10 g all the way up to the point of contact break. That would give us the maximum payload limit. For the single pad test, we attached the test object to each individual pad one after the other and repeated the experiment ten times for each pad. Because of the symmetry of the four pads, we then estimated the maximum payload for a single pad using simple average. It was found to be 7.3 N (0.75 kg) which is impressive i.e. the gripper can carry up to 3 kg payload with 0.75 kg on each pad. For testing all pads, we attached the test object to the gripper in a way such that it was in contact with all the pads simultaneously. Then we repeated the same procedure as described before to estimate the average value for the maximum payload for the gripper as a whole. We found it to be 25.5 N (2.6 kg), which is less than the sum of the average maximum payloads of the individual pads. This is because the object is not in uniform contact with all the pads, and also not with all the magnets of various pads. Also the stress or the pull force might not be exactly aligned to each of the pads and its respective magnets. So, each pad and each magnet inside a pad experiences different amount of force exerted by the payload. Fig. 3.16 shows the results for these experiments.
Minimum Drop Threshold As mentioned earlier, one motivation for designing this gripper was to reduce the minimum drop threshold all the way down to zero. This constraint is put by the servo mainly because for very small, light weight objects its force is not enough to push the object off the grip of the magnets. In case of heavier objects, gravity comes to rescue
61
Figure 3.16: Relationship between gripper state and the applied force (payload) for the gripper; (red) all three pads, (blue) a single pad only. State ’0’ means the payload is attached while state ’1’ means it is dropped.
Figure 3.17: Percentage of successful drops in relation to the payload. The success rate is 100% for almost all of the designed payload range.
62 thus making it easier to drop the object. So, in order to confirm whether or not our design achieved significant improvement in terms of minimum payload constraint, we applied the minimum load test. We simply removed the payload in steps (of 10 g) while verifying if the gripper was able to drop the object successfully. We came all the way down from maximum load to 10 g payload. The gripper was able to drop all objects with an accuracy of 100%. The success rate below 10g was around 70%. This is a great improvement compared to the previous design, since we improved the payload capability range on both sides.
Maximum Sustainable Slide Sudden obstacle avoidance or attitude change can be tricky when a quadrotor is carrying payload. In these situations, the object can be thrown off the gripper due to the slide effect of Coriolis force. The flexible gripping pads could somehow decrease this effect a little. However it still can cause a lot of trouble during aggressive maneuvers for instance. This becomes even worse outdoors in case of wind. Wind increases the side drag on the payload carrying quadrotor and hence Coriolis force become even stronger. We came up with an intelligent solution for the problem. Since the gripper is mounted on the gimbal plate, it keeps it level all the time as long as the quadrotor pitch and roll stays in between -30 to 30 degrees. So, as long as the quadrotor maneuvers stay in this region, the payload is necessarily safe from the slide, since the weight of the payload is always directed downwards and the gimbal never lets it get aligned with the Coriolis force which is acting along the direction of slide. Nevertheless we did an analysis for the worst case scenario; which is with the maximum payload (2.6 kg) and with an applied force in the horizontal plane. With the gripper mounted to the gimbal plate we applied a horizontal force at the attached maximum load until the object slid off the magnetic gripper. This value gives us the maximum sustainable slide force for the gripper. This test is not safe to repeat several times since it may result in damaging the gimbal servos. So we only conducted it
63 three times and estimated the average maximum sustainable slide to be 31.5 N. This high value of maximum sustainable slide indicates the usefulness of the gimbal.
Autonomous Aerial Grasping In order to demonstrate the operation of the whole system, especially the gripper working autonomously, we tested the quadrotor to grasp some colored test objects in the field. The objects were placed in the field at random positions. In order to assist the quadrotor to pick them up comfortably they were mounted on a base of height about 30 cm above the ground level. A drop zone was assigned in the field as well. The quadrotor was then allowed to fly above the field at an altitude of 3 m. Once it detected an object (the closest one), it would track and go over the object. Then it would hold its position and descend smoothly all the way to touch down to pick up the object. As soon as it would get the confirmation feedback from the push button mechanism, it would switch to the second camera and start ascending. After reaching an altitude of 3 m again, it would look for the drop zone using vision, and once located it would go there and drop the object. We carried out multiple tests with different colored objects with different shapes, sizes and weights. The experiments revealed, that the vision algorithms and the approach down towards the object needed tuning , however the gripper performed perfectly as far as the picking up and dropping of objects was concerned. We also subjected the quadrotor to aggressive maneuvers in between tests using the manual flight control. None of the objects was affected by the slide force even when the wind flow was around 27 kph. Fig. 3.18 shows the snapshots of the quadrotor while grasping a test object autonomously. An important thing to note here that we believe in the ability of the gripper to grasp objects of different size, shape, and curvature (Fig. 3.19) but our main focus and emphasis is rather on the autonomy of aerial grasping in these experiments.
64
Figure 3.18: The quadrotor autonomously grasping the object in the field: (1) object detection, (2) descent and alignment, (3) touch down and pick up, (4) ascend and hold to confirm pick up, (5) fly up to the designated altitude, and (6) a close up of the object carrying quadrotor.
Figure 3.19: The aerial grasping capabilities of the heavy duty gripper: The quadrotor grasping (1) metallic disc using all four pads, (2) metallic disc using one pad only, and (3) two (full) spray cans from curved surface, using two pads each for a can.
65
Figure 3.20: A quadrotor with an EPM gripper mounted on to the gimbal plate; the red dots are the push buttons for pick up/drop feedback.
3.5
Autonomous Aerial Grasping using EPM Based Gripper
For a comparative analysis, we mounted one of the quadrotors with a gripper made of EPMs EPMs are known for their less power consumption and high magnetic and hence lift capabilities. We wanted to have a comparison of the various aspects of the two technologies and their usefulness, especially in autonomous operation of quadrotors. Fig. 3.20 shows the EPM based gripper mounted to the quadrotor along with the push button based feedback mechanism which we implemented in order to achieve autonomous aerial grasping. The comparative study and experiments revealed very interesting features of our proposed design which can outclass EPMs in outdoor performance. Though a single EPM5 can lift up to 12 kg, but it requires to be aligned horizontally, and most importantly in contact with the target ferrous object in order to grasp it with full power. We tried to activate the EPM in air but it turned out it 5
OpenGrab - NicaDrone OpenGrab EPM v3
66
Figure 3.21: Screen-shot of the frames from the gripper camera live feed showing the real time view of the object being tracked on the ground at an altitude of (1) 75 cm, (2) 40 cm, (3) 25 cm, and (4) 15 cm. In this case, the camera loses track beyond the minimum altitude of 15 cm.
did not get activated to its full strength in the air; for its magnetic circuit to close, it requires contact (the technical term for this is flush) with the target object. In case of autonomous aerial grasping this kind of alignment and contact can be a big challenge. Since, when the quadrotor is very close to the object it loses its tracking ability and hence its practically very difficult to ensure that the surface of magnets are perfectly aligned to the object. Fig. 3.21 shows our experiment that explains this phenomenon. As shown, after the 15 cm threshold altitude, the quadrotor loses the track of the object (which in our case was a red colored metallic disc of diameter 18 cm). In this situation, mere vision is not a sufficient guarantee for the quadrotor to successfully grasp the object. In this kind of situation, there is no way to ensure that the magnets will be aligned perfectly with the target as the quadrotor descends blindly.
67 Moreover, in case of flat objects one might get lucky with well tuned UAVs, but this can never work in case of objects with irregular shape or curvature for instance. Our design however, being permanent in nature is able to grasp the object spontaneously, and no special alignment is required. A mere touch down with only one pad, at an angle as steep as 10 degrees (which is huge given the gripper is gimbal mounted) can easily enable the gripper to pick up the object. For the same tracking and vision algorithm the performance and average success rate of pick ups for our proposed gripper exceeded the EPMs by more than 10%. Also the weight of the equivalent EPM based gripper was found to be 840 g which is about 100 g more than our proposed design.
68
Chapter 4 Conclusion
4.1
Summary
In this thesis, we presented intelligent designs for two important robotic applications by employing the impressive latching capabilities of passive magnets. We provided the motivation and reasons of usefulness for each application by emphasizing on its applications in industry and for research. For usBot, the programmable self-assembly robot, we presented its design, its superior latching capabilities, and showed that it fulfills the constraints that we had initially set. We performed experiments to determine its performance indices and provided a brief comparison with other competing technologies while evaluating its strengths and weaknesses. Similarly, for both the passive magnetic grippers for aerial grasping, we explained the working principle, the design methodology, improvements and modifications over time, and finally the experimental results for various tests performed. We highlighted the high payload capability of our designs in addition to their high sturdiness to any possible slide, and their optimum performance in autonomous aerial grasping. We also compared our mechanisms with other existing techniques and backed the higher performance of our designs with experimental evidence.
69
4.2
Future Opportunities
As far as the problem of self-assembly is concerned, this thesis is just a starting point in terms of developing a fully autonomous robotic platform, that would demonstrate the programmable self-assembly. A lot of work still has to be done in terms of locomotion, communication, and customization of electronics for instance. However the start is quite encouraging and can be definitely used to build on it, to move forward. We identify the key directions of future work in this direction as following: To take usBot beyond just prototype i.e. move towards customization. (cus-
tomized MC, PCB, Steel frame) To introduce scalability; in terms of communication with the base station (wifi
based), and programming, and charging. In fact we are already working on the charging issue, where the face alignment magnets are to be used as charging connectors for mass charging of usBots. Develop self-actuation and self-locomotion in 2D. We are currently working
on magnetic levitation of the usBots over a uniform field. This explains the presence of magnets on the base of usBots as discussed earlier in chapter 2. Upgrade to 3D self-assembly (underwater, momentum driven, vistas)
For autonomous aerial grasping, we would like to extend this concept to multiple quadrotors and move towards collaborative lifting of objects. Also, we would like to do more extensive outdoor testings with multiple camera configurations on the quadrotors, in order to analyze the low altitude tracking, and grasping capabilities of the grippers.
70
REFERENCES
[1] D. Doyle and N. Dunne, “Self latching magnetic latching device,” Nov. 8 1994, uS Patent 5,362,116. [2] N. K. Sheridon, “Gyricon displays utilizing rotating elements and magnetic latching,” Nov. 14 2000, uS Patent 6,147,791. [3] C. Ligtenberg and B. Degner, “Magnetic latching mechanism,” Aug. 17 2010, uS Patent 7,775,567. [4] CERN,
“Superconducting
electro-magnets
at
large
hadron
collider,”
”https://home.cern/about/engineering”, 2017. [5] G. M. Whitesides and B. Grzybowski, “Self-assembly at all scales,” Science, vol. 295, no. 5564, pp. 2418–2421, 2002. [6] C. Anderson, G. Theraulaz, and J.-L. Deneubourg, “Self-assemblages in insect societies,” Insectes sociaux, vol. 49, no. 2, pp. 99–110, 2002. [7] H. C. Berg, “The rotary motor of bacterial flagella,” Annual review of biochemistry, vol. 72, no. 1, pp. 19–54, 2003. [8] S. Camazine, Self-organization in biological systems. Princeton University Press, 2003. [9] M. Rubenstein, A. Cornejo, and R. Nagpal, “Programmable self-assembly in a thousand-robot swarm,” Science, vol. 345, no. 6198, pp. 795–799, 2014. [10] A. S. Howe, “Self-assembling modular robotic structures,” IEEE robotics & automation magazine, vol. 14, no. 4, 2007. [11] A. Sproewitz, A. Billard, P. Dillenbourg, and A. J. Ijspeert, “Roombotsmechanical design of self-reconfiguring modular robots for adaptive furniture,” in Robotics and Automation, 2009. ICRA’09. IEEE International Conference on. IEEE, 2009, pp. 4259–4264.
71 [12] L. Subramanian and R. H. Katz, “An architecture for building self-configurable systems,” in Proceedings of the 1st ACM international symposium on Mobile ad hoc networking & computing. IEEE Press, 2000, pp. 63–73. [13] D. Pickem, M. Egerstedt, and J. S. Shamma, “A game-theoretic formulation of the homogeneous self-reconfiguration problem,” in Decision and Control (CDC), 2015 IEEE 54th Annual Conference on. IEEE, 2015, pp. 2829–2834. [14] C. Sung, J. Bern, J. Romanishin, and D. Rus, “Reconfiguration planning for pivoting cube modular robots,” in Robotics and Automation (ICRA), 2015 IEEE International Conference on. IEEE, 2015, pp. 1933–1940. [15] M. J. Fox and J. S. Shamma, “Probabilistic performance guarantees for distributed self-assembly,” IEEE Transactions on Automatic Control, vol. 60, no. 12, pp. 3180–3194, 2015. [16] D. Pickem, M. Lee, and M. Egerstedt, “The gritsbot in its natural habitat-a multi-robot testbed,” in Robotics and Automation (ICRA), 2015 IEEE International Conference on. IEEE, 2015, pp. 4062–4067. [17] M. Rubenstein, C. Ahler, and R. Nagpal, “Kilobot: A low cost scalable robot system for collective behaviors,” in Robotics and Automation (ICRA), 2012 IEEE International Conference on. IEEE, 2012, pp. 3293–3298. [18] H. Wei, Y. Chen, J. Tan, and T. Wang, “Sambot: A self-assembly modular robot system,” IEEE/ASME Transactions on Mechatronics, vol. 16, no. 4, pp. 745–757, 2011. [19] R. Groß, M. Bonani, F. Mondada, and M. Dorigo, “Autonomous self-assembly in swarm-bots,” IEEE transactions on robotics, vol. 22, no. 6, pp. 1115–1130, 2006. [20] J. W. Romanishin, K. Gilpin, and D. Rus, “M-blocks: Momentum-driven, magnetic modular robots,” in Intelligent Robots and Systems (IROS), 2013 IEEE/RSJ International Conference on. IEEE, 2013, pp. 4288–4295. [21] K. Gilpin, A. Knaian, and D. Rus, “Robot pebbles: One centimeter modules for programmable matter through self-disassembly,” in Robotics and Automation (ICRA), 2010 IEEE International Conference on. IEEE, 2010, pp. 2485–2492. [22] B. Haghighat, E. Droz, and A. Martinoli, “Lily: A miniature floating robotic platform for programmable stochastic self-assembly,” in Robotics and Automation (ICRA), 2015 IEEE International Conference on. IEEE, 2015, pp. 1941– 1948.
72 [23] H. A. Almurib, P. T. Nathan, and T. N. Kumar, “Control and path planning of quadrotor aerial vehicles for search and rescue,” in SICE Annual Conference (SICE), 2011 Proceedings of. IEEE, 2011, pp. 700–705. [24] L. Lin and M. A. Goodrich, “Uav intelligent path planning for wilderness search and rescue,” in 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2009, pp. 709–714. [25] C. C. Kessens, J. Thomas, J. P. Desai, and V. Kumar, “Versatile Aerial Grasping Using Self-Sealing Suction,” in IEEE International Conference on Robotics and Automation. Stockholm: IEEE, 2016. [26] P. E. Pounds, D. R. Bersak, and A. M. Dollar, “Practical aerial grasping of unstructured objects,” in 2011 IEEE Conference on Technologies for Practical Robot Applications. IEEE, 2011, pp. 99–104. [27] A. Jimenez-Cano, J. Martin, G. Heredia, A. Ollero, and R. Cano, “Control of an aerial robot with multi-link arm for assembly tasks,” in IEEE International Conference on Robotics and Automation (ICRA), 2013. IEEE, 2013, pp. 4916– 4921. [28] D. Mellinger, M. Shomin, N. Michael, and V. Kumar, “Cooperative grasping and transport using multiple quadrotors,” in Distributed autonomous robotic systems. Springer, 2013, pp. 545–558. [29] F. Augugliaro, S. Lupashin, M. Hamer, C. Male, M. Hehn, M. W. Mueller, J. S. Willmann, F. Gramazio, M. Kohler, and R. D’Andrea, “The flight assembled architecture installation: Cooperative construction with flying machines,” IEEE Control Systems, vol. 34, no. 4, pp. 46–64, 2014. [30] J. Thomas, G. Loianno, M. Pope, E. W. Hawkes, M. A. Estrada, H. Jiang, M. R. Cutkosky, and V. Kumar, “Planning and Control of Aggressive Maneuvers for Perching on Inclined and Vertical Surfaces,” in IDETC/CIE. Boston: ASME, 2015, pp. 1–10. [31] U. A. Fiaz, N. Toumi, and J. S. Shamma, “Passive aerial grasping of ferrous objects using permanent magnets,” in 2017 IFAC The 20th World Congress of the International Federation of Automatic Control. IFAC, 2017. [32] A.-a. Agha-mohammadi, N. K. Ure, J. P. How, and J. Vian, “Health aware stochastic planning for persistent package delivery missions using quadrotors,” in 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2014, pp. 3389–3396.
73 [33] MBZIRC, “Mbzirc robotics challenge,” ”http://www.mbzirc.com/challenge”, 2017. [34] DARPA, “Darpa robotics challenge,” ”http://www.darpa.mil”, 2015. [35] P. Xie, O. Ma, L. Zhang, and Z. Zhao, “A bio-inspired uav leg-foot mechanism for landing, grasping, and perching tasks,” in Proc. AIAA Atmospheric Flight Mechanics Conference, 2015, p. 1689. [36] K. C. Galloway, P. Polygerinos, C. J. Walsh, and R. J. Wood, “Mechanically programmable bend radius for fiber-reinforced soft actuators,” in Advanced Robotics (ICAR), 2013 16th International Conference on. IEEE, 2013, pp. 1–6. [37] M. E. Giannaccini, I. Georgilas, I. Horsfield, B. Peiris, A. Lenz, A. G. Pipe, and S. Dogramadzi, “A variable compliance, soft gripper,” Autonomous Robots, vol. 36, no. 1-2, pp. 93–107, 2014. [38] E. W. Hawkes, H. Jiang, and M. R. Cutkosky, “Three-dimensional dynamic surface grasping with dry adhesion,” The International Journal of Robotics Research, p. 0278364915584645, 2015. [39] J. W. Romanishin, K. Gilpin, S. Claici, and D. Rus, “3d m-blocks: Selfreconfiguring robots capable of locomotion via pivoting in three dimensions,” in Robotics and Automation (ICRA), 2015 IEEE International Conference on. IEEE, 2015, pp. 1925–1932.
74
APPENDICES
A
Papers Submitted and Under Preparation
• Usman A. Fiaz, N. Toumi, and Jeff S. Shamma, “Passive Aerial Grasping of Ferrous Objects Using Permanent Magnets”, Accepted in IFAC, July 2017. • Usman A. Fiaz, and Jeff S. Shamma, “Heavy Duty Passive Magnetic Gripper For Autonomous Aerial Grasping”, Submitted to IROS, September 2017. • Usman A. Fiaz, and Jeff S. Shamma, “usBot: A Passive Magnetic Latching Robot For Programmable Self-assembly”, Under Preparation.
