Creating robots that see the world like humans has been a challenge for scientists. Although computer vision has come a long way, these systems are still easy to fool. So why not give robots superhuman perception to compensate? MIT’s Fadel Adib has created a robot that uses radar waves to find its target, allowing it to see through walls.
The robot, known as RF-Grasp, has traditional cameras for object recognition. The camera is mounted on the bot’s mechanical gripper, giving it a good view of anything the hand might try to grab. However, what if the target is in a box or under something else? Radio waves can pass through the obstacle and RF-Grasp can use the reflected signal to locate its target.
To do this, Adib and his team used radio frequency tags, similar to those used to identify pets or open secure doors. The reader sends out RF pings, which power and modulate the tag circuitry. The reflected signal may provide data, but in this case it is used to track the physical location of the tag.
In order to test RF-Grasp, the team deployed a small focused RF reader next to the robot. The reader searches for RF beacons in its field of view, then feeds this data into the robot’s computer vision algorithm. So when told to pick up an object it cannot see, RF-Grasp relies on RF pings to find the target. When it discovers the object, the robot is smart enough to give more weight to the camera feed in its algorithms. The team says that merging the camera and RF reader data into the bot’s decision-making was the hardest part.
Compared to robots that only have visual data, RF-Grasp has proven to be much more effective in lab tests involving picking and sorting objects. It has the ability to clear clutter from the environment to find its target, guided by RF data that tells it where to dig. For example, he may remove packing material from a box to find something at the bottom. Other robots just don’t have that extra layer of guidance.
This technology could lead to robots capable of finding objects no matter where they are hidden. You have lost your keys ? Simply turn on the RF-Grasp Mk V and it will determine which pocket of which coat they are in. A more realistic application is in the warehouse industry. Robots like Boston Dynamics’ Stretch can pick up and move heavy boxes, but only if they’re visible and of regular shape. A robot with RF sensing could sort through a messy shelf to find specific items, much like a human. We could take a step closer to eliminating human labor in these environments.