Image credit: Tara Boroushaki, MIT
For years, Superman fans have wished for X-ray vision that could allow them to see objects through walls, boxes, and bins. Researchers just might make it happen. At least for robots.
A team of researchers at the Massachusetts Institute of Technology’s Signal Kinetics research group are developing RF-Grasp, a robot that uses radio waves to find hidden or lost items. Unlike vision-based solutions which are limited by a camera’s line of sight, radio waves can penetrate walls and enclosed spaces. That allows the robot to locate items that are obstructed or obscured from vision. To work, however, the target object must have an RFID tag.
All Eyes on RF
RF-Grasp uses radio frequency technology in combination with artificial intelligence and computer vision to locate and pick up items that are not in plain sight. The technology could be used to locate a TV remote or a set of lost keys within a house, but it also is a promising technology for manufacturing. The RF-Grasp could find and collect specific screws or tools needed on the manufacturing floor. It can also help pick items in a warehouse, improving order fulfillment.
RF-Grasp has a robotic arm with an eye-in-hard camera attached to its robotic wrist. The RF reader and the camera both help determine the location of the object. The robot sends out radio waves then analyzes the electromagnetic signals to measure the distance to the item. It continually updates the RF tracking data and the visual images from the camera to locate the item. As it gets closer to the object, and when it has to grasp the item, the computer vision primarily takes over.
The MIT team developed a complex motion planning algorithm to enable the robot to assess both camera and RF data streams. "It's not just eye-hand coordination, it’s RF-eye-hand coordination,” says Tara Boroushaki, a research assistant in the Signal Kinetics Group at the MIT Media Lab.
Thinking Inside the Box
The research team notes that RF-Grasp uses an RF-visual servo controller to assess the occluded object’s location and surrounding environment to establish the best path to reach the object. It also uses RF-visual deep reinforcement learning to determine how best to uncover and extract the object. In lab tests, the RF-Grasp has a 96 percent success rate when locating and collecting objects.
The research team is working to improve the robot’s grasping capabilities. The RF-Grasp is restricted in the types of objects it can pick up and the angles at which it can grasp the objects.
The RF-based technology could give humans a hand in the warehouse. Without visual perception, robots can’t locate items in a box. The radio waves can give robots the means to identify tagged items even if they are not visible to the human eye or robotic camera. In order fulfillment, that could enable robots to verify the contents of a customer order without removing and scanning each item in the box.
- Learn more about the MIT RF-Grasp technology.