Comment Re:Yeah, I did that, ten years ago. (Score 2) 15
One of the authors here.
I like when our papers are discussed on Slashdot, but often the discussion is misleading so here's some clarification.
First, let's get one thing straight: Neither we nor you could have done this back in 2007, and it's not just better hardware that allowed this today. It's also not just about RF signals going through occlusions since it is not simply about "detecting" all RFIDs near the robot. Rather, grasping requires (1) getting the RFID location right down to centimeter-scale precision, and (2) knowing how to efficiently fuse RF and visual data to maneuver the robot toward the item and grasp it.
Two things needed to come together to make this system possible:
1) You need a way to accurately measure the location of an RFID, even in practical indoor environments with significant multipath (reflections). For many years, people used methods like received signal strength (RSS) or phase, which don't work in practical indoor setups, and certainly do not when you have many reflections from objects, a robot, table, etc. because these reflections mess up RSS/phase/etc.
2) Grasping is not just about location. You need efficient navigation and practical affordance maps and a way to fuse RF and visual information.
If you want to understand how it works, you should read:
1- The paper from the same group on time-of-flight estimation on off-the-shelf RFID tags: http://www.mit.edu/~fadel/pape...
2- The follow-up on super-resolution algorithms: http://www.mit.edu/~fadel/pape...
3- The new paper on RF-visual fusion: http://www.mit.edu/~fadel/pape...
Hope this clarifies.
PS. Modern RFIDs came out of the MIT Auto-ID lab in the late 90s. Glad you came across them in 2007.
