SciTech

Research Spotlight: Modular Robotics

By Julian Whitman, Biorobotics Lab

Robots are becoming ubiquitous in industry. But, they’ve been held back from their full potential because they are difficult to incorporate into existing procedures: it’s too expensive to rebuild an entire assembly line to add in a robot. My goal is to make robots easier to use than Legos so anyone can build up a custom robot.

I’d like to be able to carry around a bag of robot modules, like joints, links, and grippers. Then, when I know what I want a robot to do, I’d tell the computer, and it would output the blueprints for a robot. I’d build a robot, and it would automatically know how to behave, like walk away, move around objects, or fasten screws.

Right now, I am focusing on automatically generating the design of robot arms specialized for a given job. For instance, I’m working with NASA to help design arms for future extraplanetary rovers. My algorithm synthesizes arms iteratively: it looks at how well the current robot can achieve its task, such as reaching a set position or exerting forces on the environment. Then, using the robot’s equations of motion, the algorithm calculates what change in the current design would help the robot complete its task more efficiently. I look for robots that have fewer joints: while robots with fewer joints may be less capable of doing arbitrary tasks than robots with many joints, robots with fewer joints are lighter, cheaper, and easier to control. In time, I hope to expand past designing a single limb at a time to designing entire robots.

Last year, I worked on making a modular legged robot walk away after you build it, generating its behavior using a layered approach. My approach included first finding a gait (the cycle of motions the robot should go through to move in the right direction), then adding onto this a layer of compliance, which prescribes how the robot should change its motions to adapt to variations in terrain. Lastly, a higher-level human or artificial intelligence planner would tell the robot how to navigate through the world.

A new modular paradigm could transform how robots are used across many domains, so that robots are viewed as customizable tools rather than monolithic entities. A search and rescue team might look at a disaster site, and build up a custom robot to go in and help survey or deliver supplies. A space mission could have a use for many different robots, but be limited in the amount of weight they can launch, which could be solved by using a set of multi-purpose robot modules. A small business owner might need to use robots to stay competitive, but need to adapt their robots to a process that changes frequently.

Making robots readily customizable will bring costs down, while increasing the scope of what robots can do and making them accessible to new users. I’m excited to be part of a movement to democratize robotics.