Radio Frequency Identification (RFID) in Robotics

We are investigating UHF Radio Frequency Identification (RFID) tags as a component of mobile manipulation and human-robot interaction systems and algorithms. Enhanced RFID sensing can provide sparse location information in addition to unique identification (even among visually identical objects). Our goal is to show that these enhanced capabilities, when fused with mobility and other sensing modalities such as laser range-finders, create more capable systems and algorithms.

Our existing efforts can be partitioned into two categories: far-field tag localization (range-bearing estimation) and near-field grasping informed by RFID.

Far-Field Localization

A particle filter that employs a physical RF sensor model and odometry can estimate the range and bearing to a tag of interest in a mobile robot's local frame. Determining range and bearing in the local frame is significant in that it allows the system to operate sans a-priori or learned maps, which is crucial for robots operating in unstructured human environments -- where tagged objects may frequently move.

The probabilistic sensor model is based on the Friis RF model, explicitly take into account multipath, and uses the manufacturer-provided antenna directivity rather that relying on cumbersome histogram approaches that require copious training.

Near-Field Grasping Informed by RFID

The choice of UHF tags is motivated by their superior far-field read ranges (up to 6m+), which makes them desirable for many mobile applications such as object fetching. However, it is also desirable to use RFID for near-field grasping, with the caveat that we want to use the SAME UHF tag as the far-field to obviate additional RFID infrastructure for the disparate regimes. While UHF tags can be read in the near-field by the (large) far-field antennas, these antennas lack discrimination capabilities, for even at the lowest possible read powers (supported by commercial readers), all tags in the near-field will respond. This makes discriminative sensing via far-field antennas virtually impossible in this near-field (grasping) regime and motivates the design of custom near-field (less than 0.3m) antennas.

To this end, we have constructed specialized near-field antennas that can be used to inform grasping. For example, the use of these near-field RFID antennas can be used to disambiguate between three (un)tagged, visually-identical objects from a laser scan in the near-field -- as shown in the images below. The location of the correctly identified object can then be used in grasping. Knowing the identity of a grasped object with high fidelity is crucial for applications where the ramifications of misidentification are severe, such as in healthcare and medicine.

Support

These projects are currently being carried out by Travis Deyle with Prof. Matt Reynolds and Prof. Charlie Kemp. Our work is generously supported in part by the Health Systems Institute and by Travis' NSF Graduate Research Fellowship (GRFP).

Publications

tag:rfid