MIT RoboGrammar System Helps Choose The Right Shape For Robots

There are several ways to figure out which robot design is the most efficient for crossing various terrains. The most time-consuming would be to simply build every variation of a robot and test them in the real world. MIT has developed a system called RoboGrammar that can test various robot designs virtually and determine which is the best for crossing particular types of terrain.

For starters, researchers tell the computer system the robot parts they have lying around, including things like wheels, joints, and so on. Researchers put in the type of terrain the robot will need to navigate, and RoboGrammar does the rest. The computer system generates an optimized structure along with the program needed to control the robot.

The lead author of the research paper is an MIT Ph.D. student called Allen Zhao. He says that robot design is a "very manual process," noting that RoboGrammar is a way to develop "new and more inventive robot designs." Zhao says that typically human-robot designers jump to robots such as quadrupeds when they need to cross various terrains, but the researchers wondered if that's the optimal design.

RoboGrammar was built as a computer model to design robots because the system wasn't unduly influenced by prior convention. While inventiveness was desired by the computer system, there were some rules in place. The team developed what they call "graphic grammar," saying that if the computer system simply connected parts and arbitrary ways, the result was a jumble of parts.

The rules were simple included things like adjoining leg segments had to be connected with a joint, not another leg segment. The rules were meant to ensure the computer-generated designs would work on a basic level. Rules controlling the system were inspired by arthropods in particular.