Many researchers conjecture module miniaturization will open the door to new ap- plications and increase the functionality of state of the art modular robots. Smaller module size enables access to the human body [53], inspection of hazardous or narrow environments [182], higher spatial resolution (e.g. for grasping) [121, 82], microma- nipulation [182], covert surveillance [103], accurate 3D physical rendering [45], and programmable matter [186, 146, 112, 41]. Indeed, early studies of modular robotics [40] recognized that modularity, unit simplicity, and decentralized control facilitate the development of microrobots.
However, module miniaturization is challenging. Fabrication and actuation of miniature modules require unconventional methods as macroscale methods become ineffective. Manufacturing thousands of small modules requires batch fabrication for
cost and time feasibility. The power to weight ratio of an electromagnetic motor, a typical module actuator, decreases significantly with miniaturization. Prominent trends in the study of module miniaturization include: active material for actu- ation, external forces for module transport, and collective actuation for scalable force output. Table 2.1 summarizes efforts towards module miniaturization of self- reconfigurable modular systems.
2.2.1
Solid State Actuation
A module typically has two types of actuators: abond actuator to provide a reversible connection to a neighboring module and a reconfiguration actuator to relocate the module during self-reconfiguration. Solid state actuation using active materials pro- vides a compact, low part count alternative to traditional electromagnetic motor actuation.
Many groups use an active material to reversibly bond modules. Magnetics pro- vide self-alignment and attraction from a distance. For example, electromagnetic bonds [76, 160] can disengage the bond without moving parts, yet have lower force to weight ratio than electro-permanent magnets [41]. Permanent magnets also pro- vide high bond force to weight ratio, but require another actuator such as a motor [160, 7, 42] or shape memory alloy (SMA) [140, 98] to break the bond. Researchers also use SMAs to actuate the latch of a mechanical bond [37, 173, 133]. Low melt- ing point alloy bonds provide a fused metal bond activated by internal [104] heat source. Other researchers use external energy for fusing modules together with low melting point alloy, but the bond is not reversible [46]. Modules in stochastic fluidic self-assembly systems (discussed below) control the flow rate through module faces using heaters to change the local viscosity of the thermorheological fluid [78, 104].
Researchers also use solid state actuation for module mobility or reconfiguration. Hawkes et al. [55] used shape memory alloy films to fold a tessellated sheet into a three dimensional shape. Ishihara et al. [64], working towards a microrobotic
CEBOT, demonstrated two types 10mm of mobile microrobots: one that used electromagnetics and one that used piezoelectrics.
There are several systems where both the bond and the reconfiguration actua- tors use active materials. In theses systems, the force used to make a new bond also transports the module. The Fracta [96] and planar Catoms [69] modules used elec- tromagnets to roll around a neighbor module during reconfiguration. The spatial, spherical Catoms module will use electrostatics to roll around a neighbor; Karagozler et al. [68] demonstrated rolling of a 1mm diameter tube using electrostatic forces. Yoshida et al. [182, 183] used shape memory alloy coil springs to actuate the mechan- ical latch bond and drive a±90◦ rotational actuator for lattice style reconfiguration.
2.2.2
External Forces
As the reconfiguration actuator typically comprises a majority of the size, weight and cost of a module, researchers study methods for using external forces to transport modules during self-assembly. Indeed structures that form by self-assembly are found many places in nature, spanning many length scales [166].
In robotic reconfigurable self-assembly, modules float randomly in an agitated Brownian environment. Modules form structure when a module randomly collides with and bonds to another module in the structure and if it desires it can break the connection. Typical bonding methods include magnetic [160, 7] and fluid pres- sure [159]. Using this principle, researchers have achieved further miniaturization of reconfigurable self-assembled robotic systems in 2D (sub-millimeter) [145, 146] and in 3D (centimeter) [104, 147]. This method of stochastic self-assembly also enables self-replication [48].
Researchers have demonstrated self-assembly at smaller length scales (from mil- limeters to nanometers), however, the processes form shapes that are not self- reconfigurable, susceptible to defects, and not reprogrammable as assembly instruc- tions are permanently set at fabrication time. Further, the process typically forms
regular, crystalline structures with limited complexity. Researchers encode assembly instructions by patterning the faces of each unit such that it mates with another unit.
At these scales, self-assembly researchers study many methods for transporting and bonding units. Units in these systems move randomly [57, 46] or guided by a rail [21] often in a fluid. Bonding force methods include low melting point alloy [46, 107, 47], capillary [8, 144, 125, 23, 22], surface tension [57], and chemical [168, 126]. While the modules in self-assembled systems move stochastically in the environ- ment before bonding to the structure, other systems use external forces to move modules deterministically. Pawashe et al. used a magnetic field and an electrostatic floor to assemble and disassemble passive magnetic tiles [112]. Gilpin et al. demon- strated self-assembly by disassembly using the force of gravity to form shape from an initial bulk of modules [42]. And Donald et al. showed that microelectromechan- ical systems (MEMS) fabricated microrobots can independently move on an active substrate and form desired shape [29].
2.2.3
Collective Actuation
In addition to exploiting configurations that best suit a task (e.g. loop for locomotion on flat terrain; hexapod for uneven terrain), modules can cooperate in parallel to increase force or torque output of the robot. Fukuda et al. [37] proposed that modules acting parallel could increase the torque output of the modular robot. Yim et al. [172] demonstrated that Polybot modules configured in a closed chain (Figure 2.5a) can exploit joint singularities to maintain high mechanical advantage over a large range of motion. Using the Deformatron system, Støy showed that modules with prismatic actuators (Figure 2.5b) can actuate in parallel to lift a load heavier than a single module’s force limit. Campbell et al. [12] presented the concept of collective actuation (Figure 2.5c) where many curved units in a lattice style system can roll on neighbors to deform the entire robotic structure.
(a) (b) (c)
Figure 2.5: Collective actuation: (a) closed Polybot chain configurations exploit joint singularities [172] ( c 2001 IEEE), (b) Deformatron modules act in parallel to increase load capacity, [137] ( c 2006 IEEE), and (c) a collective actuation cell of eight units that roll to achieve linear motion [12].