Mechanical engineering brainiacs at Cornell University and the University of Pennsylvania in the US have demonstrated how autonomous, modular robots can reconfigure themselves to accomplish specific tasks in an unknown place.
In a paper published Wednesday in Science Robotics, Mark Campbell, Jonathan Daudelin, Gangyuan Jing, and Hadas Kress-Gazit (Cornell), with Tarik Tosun and Mark Yim (UPenn) describe the system architecture, software, and hardware they used to create a self-configuring modular robot that can carry out missions in previously unmapped environments.
The project makes use of a modular robot system called SMORES-EP, which stands for Self-Assembling MOdular Robot for Extreme Shapeshifting, with EP referring to the Electro-Permanent magnets that enable module connection.
Fiction meets fact
SMORES-EP represents a small step toward realizing the dream of robots with the shape-changing capacity of fiction's Transformers, though Hadas Kress-Gazit, associate professor in the Sibley School of Mechanical and Aerospace Engineering, said, if anything, her inspiration for the project was Disney's Big Hero 6. After The Register let slip our lack of enthusiasm for the Transformer films, Kress-Gazit admitted to similar ennui.
Building on the bot boffins' prior work, the paper focuses on getting the modular robot to adapt to its environment through motion planning and sensor feedback.
"The strength of modular robots and our contribution in this paper regarding autonomous execution in unknown environments results in a robotic system that is extremely flexible and that can adjust its shape and control depending on the task and on the environment," explained Kress-Gazit in an email to The Register.
"These strengths become more pronounced in domains that are unstructured and that require the robot to perform tasks that might rapidly change."
She suggests planetary exploration and disaster relief would be appropriate applications for the technology, once it's more mature.
Using Simultaneous Localization and Mapping (SLAM) software, the robot constructs a model of its surroundings, creating waypoints for exploration and identifying environmental features that relate to its assigned task.A video posted to YouTube hints at the possibilities, so you can see for yourself how it works.
"Based on the task requirements and observed environment, the high-level planner selects an appropriate behavior for the robot from the Library," the paper explains.
"If the current configuration of the robot cannot execute the desired behavior, the high-level planner will command the Reconfiguration subsystem to transform the robot into a configuration that can execute the behavior."
Shapeshift while you work
Kress-Gazit said that the robot's task involved looking for green and pink objects and bringing them back to the zone denoted with blue. The researchers, she said, did not specify the location of the objects, how the robot should find them, or how to find the drop-off zone. Instead, the robot's behavior was informed by the data it gathered during its exploration.
The video shows how the robot's cube-shaped modules rearrange themselves into what the researchers call the Proboscis configuration – a linear, snout-like arrangement – in order to retrieve an object located in a narrow gap between two other objects. The bot's library also contains Car, Scorpion and Snake configurations.
Self-forming liquid metal just like a TERMINATOR emerges from China labREAD MORE
"The library of shapes and action (what we refer to as configurations and gaits) encodes in it assumptions about the robot's environment (for example, stair height) and limitations on the robots abilities (for example it can only move forward but not turn in a specific shape)," said Kress-Gazit. "Since we are composing the behavior, automatically, from this library, the robot will not attempt to do something it cannot."
The researchers created the library using a Unity-based tool they devised called VSPARC (Verification, Simulation, Programming And Robot Construction), which allows the construction and testing of modular robot designs and behaviors, and the importation of said code to SMORES-EP hardware.
Among the advances in this stage of the research project is the implementation of parametric behaviors. Where a static behavior might allow a particular maneuver that gets repeated the same way every time, parametric behavior considers parameters, such as turning angle and driving velocity, for a greater range of motion.
Looking ahead, Kress-Gazit said she just received funding from the National Science Foundation for a new project that will attempt to synthesize both the control and shape of a robot given a set of modules and a task.
"Another interesting direction is how we can automate the creation of the library (the shapes and controls)," she said. "Currently the library components are designed by people, but we can harness different optimization and machine learning approaches to create new candidate shapes and controls." ®