So many tasks within a house can be reduced to a series of somewhat simple movements that are repeated each time that task is done, thus making it a prime target for automation. To make this process far easier than the traditional one of designing a robot by hand, writing some code and doing thorough testing, a team of researchers from UCLA and Texas A&M has created the Mobiot toolkit, which aims to combine each of these steps into a very straightforward application that takes care of the heavy lifting automatically.
A user begins the process by utilizing their phone to first scan the target object, such as a trashcan, and then moving around the environment to simulate what motions the robot would need to do. From here, a pair of machine learning models interpret these motions and come up with a path containing a series of movements, including lifting, rotating, and simply moving forward. Once the user is confident in the virtual result, they can tell the system to transform the path into downloadable 3D parts, a list of electronics, and code along with complete assembly instructions.
By using Mobiot, anyone can now build a robot for their specific needs and have one ready-to-go in a short amount of time with minimal prior experience in robotics. For more details, be sure to read the team’s paper here and watch their CHI ’22 video below.