Chain-type robots such as CKbot look like snakes or trees, and are made of a series of connected modules. “In lattice-based robots, modules are arranged in a regular pattern; common architectures include square or hexagonal 2D shapes and cubes or dodecahedrons,” Pickem told us. A robot either rearranges or adds modules to form a shape, the self-assembly approach, or removes modules it doesn’t need, the self-disassembly approach. Hybrids such as SuperBot, created to aid NASA in planetary exploration, combine chain-type with lattice-based implementations. These assemble modules to form linear shapes or fold up to form solid shapes.
Pickem says he’s only seen prototypes so far, not finished products, of these three types. On the software side of modular robot development, several algorithms have been proven to produce a specific reconfiguration sequence. The focus has shifted recently from simulation to building hardware prototypes. The three main hardware challenges are actuation, connectors, and structural stability.
The 3D brick approach to self-assembly at the nanoscale is based on short synthetic strands of DNA that form building blocks, which self-assemble into 100 different, precise 3D shapes such as letters and numbers. Like the models of 80 of these shapes shown here, each unique shape measures about 25 nm per side.
“Actuation challenges include the fact that robots must be small, yet strong enough to lift themselves and other modules,” Pickem said. “Connectors must form reliable and strong connections for structural stability, yet also break when necessary. Because modular robots don’t have the same structural stability as monolithic robots designed for the manufacturing floor, the challenge there is how to make them both light and small, as well as strong.”
Self-reconfigurable robots can be built using different design principles: deterministic or stochastic, self-assembling or self-disassembling, centralized or decentralized, and homogeneous or heterogeneous. Deterministic schemes can locate modules at any given time, but require more planning and control because they tell every module what to do. In stochastic architectures, modules’ connections and disconnections happen randomly, and are more likely to occur as module count increases.
Self-assembly schemes are more common than self-disassembly schemes, said Pickem. One self-disassembly method has been built by a team led by Daniela Rus, a principal investigator at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL). Developed under the aegis of CSAIL’s Distributed Robotics Laboratory (DRL), small robotic cubes self-disassemble to duplicate an object placed in a heap of them. Measuring 12 mm per side, the Smart Pebble robotic cubes communicate how to align themselves to duplicate an object’s shape using distributed algorithms. First, they form a grid using electropermanent magnets, then they discard unneeded cubes. The team wants to scale cube size down to 1 mm per side to build thousands of cubes on a silicon wafer using lithography.
In a centralized architecture one agent plans for all modules, but in a decentralized scheme every module plans for itself based on information it observes or gathers by communicating with neighbors. Generally, large numbers of robots can be controlled more efficiently with decentralized control schemes.
In a homogeneous architecture, modules have the same properties, so are interchangeable and can be replaced easily, a more robust design than specialized, fixed-architecture monolithic robots. Robots made of heterogeneous modules can be more flexible in their capabilities. “One can think of a mobile robot with dedicated battery, wheel, actuation, or sensor modules, which has all the capabilities of its individual modules,” said Pickem. “Overall functionality is improved at a much lower cost. To extend a homogeneous robot, all modules must be extended with the desired capability.”