A robot that can tidy up a room without outside help
Posted: Thu Dec 12, 2024 9:28 am
Robots are great at some tasks, like lifting and moving objects, and even cooking, but getting them to do all of this in an unfamiliar environment outside of a lab is a real challenge.
A new system called OK-Robot can fix this, which allows you to quickly teach robots to navigate where they've never been before.
To develop the OK-Robot system, researchers from New York Bulk SMS Australia University and Meta tested Hello Robot's Stretch robot in 10 rooms across five homes. Stretch consists of a wheeled module, a tall pole, and a retractable "arm."
The scientist scanned the space using Record3D, an iPhone app that shoots 3D video using the phone's lidar system and transmits its work.
The OK-Robot system then reviewed the video footage using artificial intelligence to detect objects. Combined with other models, this helped the robot identify objects in the room, such as a toy dragon, a tube of toothpaste, and a deck of playing cards, as well as a chair, table, and trash can.
The team then instructed the robot to pick up an object and move it to another location. Stretch did this 58.5% of the time, and 82% of the time in less cluttered rooms (the study has not yet been peer-reviewed).
The OK-Robot system used models that were not specifically tuned for this project. So when the robot didn't find the right object, it simply stopped. This is one of the reasons why it performed better in neater rooms: the fewer objects, the less chance it had of getting tangled and the more space it had to navigate.
Ready-made open-source models have both pros and cons, says Larrel Pinto, an associate professor of computer science at New York University who was one of the project leaders.
A new system called OK-Robot can fix this, which allows you to quickly teach robots to navigate where they've never been before.
To develop the OK-Robot system, researchers from New York Bulk SMS Australia University and Meta tested Hello Robot's Stretch robot in 10 rooms across five homes. Stretch consists of a wheeled module, a tall pole, and a retractable "arm."
The scientist scanned the space using Record3D, an iPhone app that shoots 3D video using the phone's lidar system and transmits its work.
The OK-Robot system then reviewed the video footage using artificial intelligence to detect objects. Combined with other models, this helped the robot identify objects in the room, such as a toy dragon, a tube of toothpaste, and a deck of playing cards, as well as a chair, table, and trash can.
The team then instructed the robot to pick up an object and move it to another location. Stretch did this 58.5% of the time, and 82% of the time in less cluttered rooms (the study has not yet been peer-reviewed).
The OK-Robot system used models that were not specifically tuned for this project. So when the robot didn't find the right object, it simply stopped. This is one of the reasons why it performed better in neater rooms: the fewer objects, the less chance it had of getting tangled and the more space it had to navigate.
Ready-made open-source models have both pros and cons, says Larrel Pinto, an associate professor of computer science at New York University who was one of the project leaders.