Robots at The University of Texas at Dallas are learning to recognize objects through a new system developed by a team of computer scientists. The system allows the robot to push objects multiple times until a sequence of images is collected, which helps the robot recognize the objects. This is a significant advance in artificial intelligence (AI) for robots, as it helps them better identify and remember objects.
The Significance of Object Recognition for Robots
Object recognition is crucial for robots to perform tasks such as picking up items or bringing objects to humans. This technology is designed to help robots detect various objects found in different environments, like homes. It can also generalize and identify similar versions of common items, even if they come in different brands, shapes, or sizes.
How Robots Learn to Recognize Objects
In the lab at The University of Texas at Dallas, researchers use a robot named Ramp to train the AI system. Ramp is a mobile manipulator robot with a long mechanical arm and a square “hand” to grasp objects. The robot learns to recognize objects by pushing them. After pushing the object multiple times, the robot collects a sequence of images, which are used to train the AI model. With this training, the robot can recognize the object without pushing it again, just like how children learn to interact with toys.
Previously, interactive perception methods only used a single push by the robot. However, this new system pushes each item 15 to 20 times, enabling the robot to capture more photos with its RGB-D camera. This detailed information reduces the chances of mistakes and improves the robot’s ability to recognize and differentiate objects.
Next Steps for the Research
The researchers at The University of Texas at Dallas are continuing to improve other functions of the robot system, including planning and control. These advancements could enable robots to perform tasks like sorting recycled materials. The team hopes to expand the capabilities of robots and make them more versatile in completing complex physical tasks.
The research was supported by the Defense Advanced Research Projects Agency, which aims to develop AI technologies that assist users in performing physical tasks and reducing errors.