Researchers have developed a robot to find the lost objects

Oct, 2021 - By SMI

The robotic arm uses the data from an antenna and camera to fetch things even deep beneath piles.

MIT researchers have developed a robotic system that can perform this. The RFusion system consists of a robotic arm with a camera and an RF antenna connected to its grip. It combines antenna signals with visual information from the camera to identify and retrieve an object, even if it is buried behind a mound and entirely hidden. Affordable RFID tags battery-free tags that can be attached to an item and reflect signals supplied by an antenna are used in the RFusion prototype built by the researchers. RFusion can find a tagged object within a pile since RF waves can pass over most surfaces.

The robotic arm uses machine learning to instantly zero in with the precise target object, move the objects on top of it, hold the object, and verify that it has picked up the correct item. Since the camera, antenna, robotic arm, and artificial intelligence are all fully integrated, RFusion could operate in any surroundings without requiring any specific setup.

Reinforcement learning was used to train a neural network that can optimize the robot's path to the target. The algorithm is learned by trial and error with a reward mechanism in reinforcement learning. Once the system has found the precise correct place, the neural network utilizes a combination of RF and optical data to anticipate how the robotic arm should grab the object, including the angle of the hand and grip width, as well as if other things must be removed first. It also does the last scan of the item's tag to ensure that it has picked up the correct item. This strategy worked nicely. When it came to recovering things that were completely buried behind a heap, RFusion achieved a 96 % success rate.

Even as locating lost keys is useful, RFusion could have a variety of other uses in the future, such as sorting through piles to process orders in a storage facility, classifying and installing components in an auto manufacturing facility, or assisting an old person with everyday tasks in the residence, however, the prototype model isn't fast enough for some applications yet.