1. An RFID-Based Object Localization Framework and System
Kirti Chawla
Department of Computer Science
University of Virginia
Good morning, my name is Kirti and today I am going to present my PhD dissertation talk on – An RFID-Based Object Localization Framework and System

2. Location, Location, Location
Introduction
Locate
Objects
Environments
Goal: Locate objects in an environment
Attributes:
Reliable
Accurate (sub-meter) and Fast (seconds)
I am addressing the problem of locating objects within an environment in a reliable, accurate and fast manner. Locating objects is an important problem to solve because it occurs in a variety of settings such as locating milk cartons in a warehouse, providing users with location-based insights, etc.

3. Far-field Communication Near-field Communication
RFID Primer
Background
RFID Reader
RFID Tag
Far-field Communication
Near-field Communication
Tags and Readers:
Form Factors
Operating Frequency
Power Source
In this research, I’m using an automatic identification technology called RFID for locating objects in indoor environments such as warehouses, hospitals, airports, etc. Before we go further, lets look at the basics of this technology. RFID technology consists of two components – a reader and a tag. Both of these devices can communicate with each other using two different mechanisms – near-field and far-field communication.
In near-field communication, tag and reader communicate using electro-magnetic induction coupling of their coils over a short distance. In far-field communication, tag and reader communicate using radio signal backscattering over a long distance.
Tags and readers come in a variety of form factors and can operate over a wide-range of frequencies and power sources.

4. Intellectual Contributions
Adaptability
 Resilient to environmental conditions / noise
Accommodates numerous scenarios
 Tag orientation and vendor hardware –agnostic
Reliability
 Signal strength as a reliable metric
 Tag sensitivity influences performance
 Tag selection & sorting ensures uniformity
 Heuristics enhance accuracy
Objects are located using RFID technology by attaching RFID tags to them.
I have made several key contributions that impact the adaptability, reliability, and scalability of the RFID-based object localization. For example, my approach can locate arbitrarily oriented tags using different RFID readers.
To enhance reliability, my approach selects and sorts the tags.
For enabling scalability, my approach makes reference tags optional without sacrificing accuracy.
Scalability
 Tag selection optimizes range & cost
 Improved performance by matching tags to readers
 Reference tags are unnecessary

5. Current State of the Art
Background
Technologies
Mismatched Solutions
Techniques
Locating objects is an important problem that has been studied extensively from a variety of technologies, techniques and use-cases standpoint. For example, GPS is a location system that uses geo-stationary satellites for locating vehicles. Often, these technologies, techniques, and use-cases are combined in an ad-hoc manner leading to mismatched solutions with limiting constraints. For example, while using GPS to locate objects in an outdoor setting makes perfect sense, the same technology cannot be used in indoor environments due to unavailability of signals.
Limiting Constraints

6. Pros/Cons Motivation Pros Cons No Line of Sight Solid Obstacles
Unintended Use
Invasive
Dark Environment
Susceptible
Lets look at the pros and cons of using RFID technology for locating objects. On the pros side, RFID can locate objects beyond line of sight, through solid obstacles and in dimly lit environments. Moreover, passive tags cost a few pennies each and enable cost-effective localization. RFID is adaptive to various application requirements.
On the cons side, RFID was not invented to locate objects and thus, its harder than usual to derive an unintended functionality. RFID is perceived as an invasive technology that intrudes privacy. Signal strength based RFID localization approaches are susceptible to environmental interferences and occlusions.
RFID has a high entry barrier as it requires new hardware to be deployed for locating objects. RFID can only work for specific use-cases and is not a panacea for all localization problems.
Cost
Effective
Adaptive
Entry Barrier
Targeted

7. Potential New Savings = $ 600 Million / Year
Use-Case: Warehouse
Motivation
100, 000 Ft2
4000 Stores
Floor space and Nos.
Warehouse-Store
100 People
$ 12/Hour
275 Days/Year
Workforce Cost
30 Min./Day
Avg. Search Time
Lets look at a few use-cases where RFID-based localization may be useful. Consider a large-scale warehouse operated by Walmart. There are 4000 such warehouses in US each with a floor-space of 100,000 sq-feet. To manage such a space, about 100 people work for 275 days per year at the cost of 12 dollars per hour. These people spend about 30 minutes per day finding items. Cumulatively, this amounts to 600 million dollars worth of wasted annual productivity.
If these items could be found sooner then this loss could be turned into potential new savings.
Potential New Savings = $ 600 Million / Year

8. Other Use-Cases Motivation
Locate:
Medical Supplies
Surgical Instruments
Caregivers
Patients
Hospitals
Locate:
Guests / Travelers
Freight
Baggage
Airports
Other potential application scenarios of RFID-based localization include hospitals for locating medical supplies, surgical instruments, caregivers, and patients and airports for locating guests and travelers, freight, and baggage.

9. Thesis Statement Research
Performance Enhancing Heuristics
Empirical Power-Distance Relationships
Uniformly Sensitive Tags
Reliable High-Performance RFID-based Object Localization Framework and System
Here’s my thesis statement: a reliable and high-performance RFID-based object localization framework and system can be developed using uniformly sensitive tags, empirical power-distance relationships, and performance-enhancing heuristics. In the following slide I will describe how different parts of this thesis statement is realized in practice.

10. Localization Framework
Research
Collection of Tags
Tag Selection
Candidate Tags
Tag Binning
Uniformly Sensitive Tags
Empirical Power-Distance Relationship
To address the problem of locating objects using RFID, I’ve developed a localization framework comprising of four stages – i.e., Tag selection, Tag binning, Empirical power-distance relationship, and Performance-enhancing heuristics. The first two stages ensure that tags used for localization purposes are uniformly sensitive while third and fourth stage focus on tags’ location estimates and improving those estimates, respectively. This is how the thesis statement mentioned in the previous slide is realized in practice. In the following slides, I will briefly describe each of these stages.
Tags’ Location Estimates
Performance-Enhancing Heuristics
Improved Location Estimates

11. Mobile Robot with onboard reader and multi-tag
Experimental Setup
Evaluation
RFID Reader
Backend Host
Tablet
Internet
Antenna
Reference Tag
**** show robot and tablet ****
Here is an overview of the RFID lab where the localization experiments were conducted. I have developed a track-based robot system for simulating stationary and mobile objects. Also, reader antennas were placed on the sides and the ceiling as shown here. The localization algorithms, models, and heuristics run on the backend host that collects empirical power-distance relationship data to locate the target tags. These location estimates are then forwarded to modern devices such as tablets for visualization purposes.
I will now focus on the four stages of the localization framework.
Mobile Robot with onboard reader and multi-tag

12. Tag Selection Research
Tag Collection
Read Range
RSS
Read Count
Tag Selection
Candidate Tags
Key Contribution:
Tag Selection Process
**** show the pile of tags *****
Consider this, different tags can have different performances and yet state-of-the-art RFID localization approaches do not take this into account. This is a critical issue because under-performing tags will cause more readers to be deployed thereby raising the overall solution cost.
To mitigate this issue, In the tag selection stage we select the best performing tags from a collection of different tags using read range, received signal strength and read count metrics.
Read range is defined as the longest distance a tag can be read from the reader, received signal strength or RSS in short, is the amount of reflected power a reader receives from the tag, and, read count is the number of times a tag is read by the reader in a given time.
On a side note, the blue arrow, such as shown here, shows the results of tag selection stage and such markers are used across the talk to show relevant results.
Problem: Tags have variable performance
Solution: Select tags based on their performance

13. Tag Selection Evaluation
Insight: Select tags on multi-objective criteria
Lets look at a result related to the tag selection stage.
In this experiment, I take several tags of different types and measure their performance over different power-distance combinations. Here is a graph that shows tag types on the x-axis and their average RSS performance on y-axis. It is evident that tag-14 is best performing tag as it performs consistently over different power-distance combinations.

14. Same Type Tags Collection Uniformly Sensitive Tags
Tag Binning
Research
Same Type Tags Collection
RSS
Read Count
Tag Binning
Uniformly Sensitive Tags
Key Contribution:
Tag Binning Process
*** show two large tags of same type ****
Furthermore, state-of-the-art RFID localization approaches assume that tags of same type have similar sensitivities, which is not true. I have discovered that two tags of same type can have vastly different sensitivities due to manufacturing variation. This critical issue impacts the performance of signal strength based RFID localization approaches.
To mitigate this issue, In the tag binning stage we sorts the same type tags based on their sensitivity using received signal strength and read count metrics. The result is a Gaussian distribution of tags from which I pick tags about the mean to arrive at uniformly sensitive tags.
These uniformly sensitive tags are used in my localization experiments.
Problem: Tags have variable sensitivities
Solution: Bin tags based on their sensitivity

15. Tag Binning Evaluation
Insight: Sort tags on their RF performance
0.61 meters
1.83 meters
3.05 meters
Here’s a result from the tag binning stage.
In this experiment, I take 500 tags of tag-14 type and sort them according to their sensitivities. Here is a graph that shows average RSS behavior on the x-axis and number of tags on the y-axis.
Colored arrows indicate mean of the Gaussian distributions at different distance levels. I pickup tags about these means and compute an intersection to get the final set of uniformly sensitive tags.

16. Tags are uniformly sensitive
Tag Binning
Evaluation
0.61 meters
1.83 meters
3.05 meters
Yield: ~70 % (350 out of 500)
Tags are uniformly sensitive
In essence we get uniformly sensitive tags from different Gaussian distributions and I compute intersection over these sets to arrive at the final set of uniformly sensitive tags. For the tag-14 case, about 70% or 350 out of 500 tags were found to be uniformly sensitive.

17. Friis Transmission Equation
Power-Distance Relationship
Research
Friis Transmission Equation
Transmitted Power: PT
RFID Reader
Tag-Reader Distance: D
Received Power: PR
Before we go deeper into the third stage of the localization framework, lets look at the Friis transmission equation that theoretically estimates tag-reader distance using signal strength.
This estimate is highly inaccurate in real-world scenarios due to various interferences and occlusions. Thus, empirical power-distance relationship is needed for higher localization accuracy.
RFID Tag
Problem: RF signal variability renders Friis Eq. useless
Solution: Utilize empirical power-distance relationship

18. Environment Dependent
Power-Distance Relationship
Evaluation
Insight: Empirical power-distance relationship enables higher performance
Ideal Friis (N = 2)
Ideal Friis (N = 3)
Ideal Friis (N = 6)
Empirical
Environment Dependent
Here’s a result that shows the difference between theoretical and empirical power-distance relationship.
In this experiment, I theoretically and empirically measure the reader transmitted signal strength needed to detect a tag. Here is a graph shows tag-reader distance on the x-axis and transmitted signal strength needed to detect the tag on the y-axis. Black lines indicate ideal Friis-based estimation using different decay rates, while the red line shows empirical estimation. It is evident that empirical power-distance relationship enables higher accuracy.

19. Power-Distance Relationship Research
Read Count
Empirical Power-Distance Relationship
TX-Side Algorithms
RX-Side Models
Uniformly Sensitive Tags
Tags’ Location Estimates
In essence empirical power-distance relationship involves experimentally determining variation in signal strength as tag-reader distance varies. Here, I take this relationship and split it into two complimentary halves – i.e., transmitting side and receiving side. Consequently, I develop transmitting-side algorithms and receiving-side models that utilize these relationships to help locate objects. I will now focus on transmitting-side algorithms.
Problem: Locate objects using empirical power-distance relationship
Solution: Utilize TX and RX empirical power-distance relationship

20. Locate Tags: Power-Modulating Algorithms
TX-Side Algorithms
Research
Locate Tags: Power-Modulating Algorithms
Radio Wave
Shared Region
Since we use uniformly sensitive tags, a key insight is that similarly behaving tags are neighbors. To visualize this, consider a region with RFID reader antennas and two uniformly sensitive tags that are neighbors. I algorithmically determine the minimum transmission signal strength needed to detect these tags and repeat the process from different sides. As the two tags are neighbors they are detected at similar signal strengths. Alternatively, if they are detected at similar signal strengths then they must be neighbors.
Antenna
Insight: Similarly behaving tags are neighbors

21. Locate Tags: Power-Modulating Algorithms
TX-Side Algorithms
Research
Algorithms
Locate Tags: Power-Modulating Algorithms
Key Contributions:
TX-Side Power-Modulating Algorithms
I utilize this insight to develop algorithms for locating the tags. In particular, I layout a region with reference tags at known locations and empirically establish their power-distance relationship. Consequently, I put a target tag at unknown location and repeat the process. I then correlate its empirical power-distance relationship with that of reference tags’ to arrive at its position estimate.
Problem: Locate tags using TX RF signal power
Solution: Algorithmically modulate TX RF signal power

22. Overall Accuracy: 0.18 meters
TX-Side Localization Accuracy
Evaluation
Insight: Performance can be improved by denser reference tag deployment
Time
Overall Accuracy: 0.18 meters
Here’s a result of localization accuracy using transmitting –side algorithms.
In this experiment, I place a mobile robot having an onboard tag on the track and locate it using the algorithms.
Here is a graph that shows different measurement points along the robot’s path on x-axis and average distance on
the y-axis. It is evident here that the inferred location estimates closely follow the robot’s actual locations, except at
a few places where the reference tags density was not sufficient enough.

23. Density Vs Performance
Evaluation
Insight: Localization performance varies with reference tag density
Another key result pertaining to TX-side algorithms is showing that reference tags can be counter-productive.
In this experiment I use transmitting-side algorithms to measure the localization accuracy while gradually increasing the reference tag density.
While reference tags do improve the accuracy, such improvements decrease as the density increases and stops beyond a point of diminishing returns.
Using more reference tags beyond that point is counter-productive as the cost of the overall solution increases without significantly improving the accuracy.

24. Tag-Reader Matched Pairs
RX-Side Models
Research
RFID Reader - A
RFID Tag - A
Key Contributions:
Tag-Reader Matched Pairs
RFID Reader - B
RFID Tag - B
When you have several different tags and readers, its natural that some of them will perform better than the other. Thus, the key insight here is that select tags can be matched to readers using read-range, received signal strength, and read-count metrics for providing higher accuracy.
Insight: Match tags to readers for higher performance

25. Locate Tags: RSS Decay Models
RX-Side Models
Research
Locate Tags: RSS Decay Models
Friis Physics Model
RSS Decay Model
Key Contributions:
Tag Orientation Inclusive
RSS Decay Models
Radial Orientation
Axial Orientation
I use this key insight to adapt Friis theoretical model to reality. In particular, I model received signal strength decay of the matched tag-reader set with respect to its tag-reader distance.
While making these models, I take into account tag’s axial and radial orientation because tag orientation impacts the accuracy. Thus, if models are orientation –agnostic then the objects can be located in an orientation-free manner.
Moreover, such model based localization approach does not need reference tags for providing higher accuracy. But, reference tags can still be used to further improve it.
Problem: Locate tags using RX RF signal power
Solution: Adapt theoretical physics model to reality

26. RSS Decay Models Evaluation
Insight: Orientation-based decay models lead to orientation-agnostic localization
Radial
Here we see a result of modeling received signal strength decay over tag-reader distance.
In this experiment, tag-reader distance is gradually increased while measuring the received signal strength. Here is a graph that shows
tag-reader distance on the x-axis, average RSS behavior on the y-axis, and tag’s axial orientation on the z-axis. An important observation here
is that received signal strength decay is smooth because we’ve used tags selected through tag selection and binning stages.
Also, modeling the decay with respect to tag orientation ensures orientation-free localization.

27. RX-Side Localization Accuracy Evaluation
Insight: Performance can be improved by minimizing RF dead-zones
Overall Accuracy:
meters
Here’s a result of localization accuracy using receiving-side models.
In this experiment, I place a mobile robot having onboard tag on the track and locate it using the models. Here is a graph that shows
different measurement points along the robot’s path on the x-axis and average distance along the y-axis. It is evident here that the
inferred location estimates closely follow the robot’s actual locations, except at a few places where the radio signals were partially
available.

28. Scalability: No. of Objects
Evaluation
Insight: No. of objects -invariant localization accuracy feasible
Here’s a result showing that my approach scales to large number of objects.
In this experiment I gradually increase the number of objects to be localized while measuring their accuracies. It is evident that
my approach provides sustained localization accuracy as the number of objects increase.

29. Heuristics for Improving Localization Accuracy
Research
Heuristics
Localization Error
Key Contributions:
Heuristics for Improving
Localization Accuracy
Target Tag
Reference Tag
In the final and optional stage, the location estimates provided by the previous stage can be further improved by carefully selecting neighbor reference tags. Towards this end, I’ve developed several heuristics that enable selection of such reference tags.
Problem: Assumption that target and reference tag location coincide leads to localization error
Solution: Consider neighbor reference tags that minimize localization error

30. Experimental Setup Evaluation
RFID Reader
Backend Host
Tablet
Internet
Antenna
Target Tag
Here’s an overview of Digital Medial Lab were scalability related experiments were conducted.

31. Scalability: Environment
Evaluation
Insight: Scale-invariant localization accuracy
feasible
Overall Accuracy: 0.32 meters
*** noise measurements show DML is noisier than Walmart ***
Here’s a result showing that my approach scales to larger environment.
In this experiment, we locate objects in Digital Media Lab having a larger area than the RFID Lab. It is
evident that the inferred location estimates closely follow the object’s actual locations.

32. Comparative Evaluation
Approach
Localization
Time
Test Region (m2)
Localization Accuracy (m)
Reference Tags
Ni et al., 2003
Not Reported
2D, 20 2
Active
Bekkali et al., 2007
2D, 9
0.5 – 1.0
Passive
Zhao et al., 2007
2D, 20
0.14 – 0.29
Choi and Lee, 2009
2D, 14
0.21
Choi et al., 2009
2D, 3
0.2 – 0.3
Zhang et al., 2010
2D, 36
0.45
Brchan et al., 2012
A few seconds
2D, 22
1-2
TX-Side:
Combined Algorithms
1.67 minutes
2D, 8
0.18
RX-Side:
Combined Models
(without ref. tags)
~4 seconds
Not Applicable
(with ref. tags)
Variable
Key Results:
Localization Accuracy (Sub-meter)
Localization Time (A few seconds)
Reference Tags (Optional)
Here’s a table that compares the localization performance of my approach with the other state-of-the-art approaches.
My approach locates objects with sub-meter accuracy and in a few seconds in noisy indoor environments. My approach does not need reference tags for providing higher accuracy. Moreover, several of these approaches use costly battery-powered active tags for localization, which is not scalable.

33. Summary and Future Work
Conclusion
RFID-Based Location System:
- Pure RFID reliably locates objects
Match tags to readers
Tag selection & binning improves tag performance
TX/RX empirical power-distance relationship
Algorithms, models, and heuristics for object localization
Identify / mitigate key localization challenges
Future Research Directions:
3D Visualization tools
Field testing and commercialization
In summary, I’ve developed an indoor GPS like location system using RFID technology that utilizes matching tag-reader pairs, uniformly sensitive tags, and empirical power-distance relationships. I’ve also identified and mitigated key localization challenges.
In the future, we plan to develop 3D visualization tools, perform in-depth field testing and focus on commercialization.

34. Deliverables Contributions
Co-directed 10 undergraduate theses and Capstone projects
Won the 2011 SEAS Entrepreneurial Concept Competition
Placed 2nd at the 2012 Darden Business Competition
Best Presentation Award at 2013 IEEE Conference on Localization
Journal Publications:
Kirti Chawla, Christopher McFarland, Gabriel Robins, and Wil Thomason,
An Accurate Real-Time RFID-Based Location System, 2014, In Submission
Kirti Chawla and Gabriel Robins, An RFID-Based Object Localization
Framework, International Journal of Radio Frequency Identification
Technology and Applications, Inderscience Publishers, 2011, Vol. 3, Nos. 1/2, pp. 2-30
Conference Publications:
Kirti Chawla, Christopher McFarland, Gabriel Robins, and Connor Shope, Real-Time RFID Localization using RSS, IEEE International Conference on Localization and Global Navigation Satellite System, 2013, Italy, pp. 1-6
Kirti Chawla, Gabriel Robins, and Liuyi Zhang, Efficient RFID-Based Mobile Object Localization, IEEE International Conference on Wireless and Mobile Computing, Networking and Communications, 2010, Canada, pp
Kirti Chawla, Gabriel Robins, and Liuyi Zhang, Object Localization using RFID, IEEE International Symposium on Wireless Pervasive Computing, 2010, Italy, pp
Patents:
Kirti Chawla and Gabriel Robins, System and Method For Real-Time RFID Localization, 2014
Kirti Chawla and Gabriel Robins, Real-Time RFID Localization Using Received Signal Strength (RSS) System and Related Method, US Patent: 61/839,617, 2013
Kirti Chawla & Gabriel Robins, Object Localization with RFID Infrastructure,
WIPO Patent: A3, 2012; US Patent: A1, 2013
Here are the accomplishments of this research. I have co-directed 10 undergraduate theses and capstone projects and won numerous entrepreneurship awards. Also, I have published in several international journals and conferences and filed numerous patents.
Here are some of my student colleagues that have actively contributed to this research. Thank you and I will be happy to take questions.

35. Object Localization with RFID Infrastructure
USPTO and WIPO
Patents
Object Localization with RFID Infrastructure

36. Backup Slides

37. Backend: Minimize Misuse Potential New Savings = $ 200 Million / Year
Motivation
Back
100, 000 Ft2
4000 Stores
Floor space and Nos.
Warehouse-Store
1 Million Items
5 % Misuse Rate
$ 1 / Item
Reported Misuse
Potential New Savings = $ 200 Million / Year

38. Frontend: Improve Turnaround Motivation
Back
100, 000 Ft2
4000 Stores
Floor space and Nos.
Warehouse-Store
$ 319B Rev/Year
$ 79M /Store/Year
$ 218K /Store/Day
Revenue Generation
$ 72 /Day/Person
3K /Store/Day
+5 /Store/Day
Maximize Utility
Potential New Revenue = $ 500 Million / Year

39. Motivation Minimize Misuse Improve Turnaround Save Time
How Our Research Can Affect Your Bottom Line
Motivation
$ 200 Million / Year
Minimize Misuse
Improve Turnaround
$ 500 Million / Year
$ 600 Million / Year
Save Time
Stimulate Spending
$ 4.3 Billion / Year

40. Localization Challenges
Approach
Radio Interference
Occlusions
Tag Sensitivity
Tag Spatiality
Tag Orientation
Reader Locality

41. Reliability through Multi-Tags
Approach
RFID Reader
Platform Side View
Platform Top View
Parallel
Orthogonal
RFID Tag
RFID Tag
Vertical
Horizontal
Problem: Optimal tag reads occur at certain orientations
Solution: Multi-Tags provide orientation redundancy

42. Reader Output Power Range
Power-Modulating Algorithms
Approach
Back
Reader Output Power Range
MAX
MID
Linear Search
Binary Search
Parallel Search
O(#Tags 
#Power-Levels)
O(#Tags  Log#Power-Levels)
O(#Power-Levels)

43. Minimum Power Selection
Heuristics Framework
Approach
Back
Absolute Difference
Minimum Power Selection
Localization Error
Meta Heuristic
Root Sum Square
Problem: There can be multiple neighbor reference tags
Solution: Select neighbor reference tags using different selection criteria

44. RSS Decay Models Evaluation
Back
Insight: Orientation-based decay models lead to orientation-agnostic localization

45. TX-Side Localization Time
Evaluation
Back
Insight: Faster algorithms provide lower tag detectability

46. Technology Cost Breakup (Post R&D)
Product
100, 000 Ft2
4000 Stores
Floor space and Nos.
Old Revenue = 79M / Store / Year
Warehouse-Store
$ 20K (300 Ant.)
$ 20K (80 Readers)
$ 10K (1M Tags)
RFID Hardware Cost
Variable (Software)
$ 50K (Backend)
Software and Misc. Cost
New Revenue = 81M / Store / Year
Total Cost 1st Year = $ 100K + SLC* + AMC+ / Store
Total Cost Nth Year = SLC + AMC / Store; N≥ 2
* Software License Cost, + Annual Maintenance Cost | All costs are current estimates

47. Locate Readers: Proximity-Sensing Algorithm
TX-Side Algorithms
Research
Locate Readers: Proximity-Sensing Algorithm
Problem: Locate readers using TX RF signal power
Solution: Sense proximity of neighbor tags

48. Ambient Noise: Frequency
Evaluation
RFID Lab (without)
RFID Lab (with)
D M Lab
Walmart