Ellie Smelly! Capture smells to build Smells Collections

It’s 2021, but sure it feels like 2020 to me. We are still struggling to find a way out of this pandemic. Desperation is the true mother of all inventions, and so in moments of need we look for any solution that can help. Currently we are all looking on how to learn to live with COVID endemic global catastrophe.

It has been suggested that dogs are particular good at detecting smells of diseases. Can we use them for detecting COVID on people? How about technology? A recent report (published after this contest started and ideas submitted) from a group of researchers from the London School of Hygiene & Tropical Medicine (LSHTM) and the biotech company RoboScientific Ltd. with Durham University (also in England) describes how, with some success, VOC sensors can be used to detect COVID signatures in the air.

Although my initial idea was just that, the lack of sample data (COVID patients) and my lack of desire to be near infected people inhibited the development of a prototype for the contest. However, I realized that more than just COVID detection there are numerous applications waiting to be developed if we have a collection of smells available.

Demonstrated by the origin of captcha, humans (and dogos like Ellie) are very good at identifying things that will be very complex for an untrained sensors or computer to do. In the case of smells, current (costly) devices are good at identifying the elements, and quantities of sensed sample. However, we humans don’t care to know what elements, amino-acids, sugars, alcohols or whatever combination produces the coffee aroma, we just know it smells like coffee. A sensor can give us a signature (a reading of TVOC/eCO2 values using Adafruit SGP30 sensor used in this prototype) that a human can map to a particular entity. A better sensor will produce additional parameters that combined with machine learning (ML) and human/dogos help, will create a more precise signature of a particular smell.

Scientist have described the characteristics of smells and method of predictions [1] [2]. Information theory has demonstrated in many ways the encoding methods. The adoption of a common sensor(s) will result in uniform encoding of smells. In addition, the captcha model can be employed to crowd-source the capture, categorization of smells, and not of just primitive aromas like coffee, but complex combinations. I know elementary schools are full of volunteers that will be willing to send good data for a badge and recognition. ML can be used to curate the user submitted data to produce a read only collection of smells. The combination of these technologies and process will result in a global smell collection.

This collection of smells can then be used to develop applications in health, security, education and even entertainment. For example:

· In health: diagnosis of diseases; food safety

· Security: explosives, forensics, and detection of drugs

· Household and building appliances: branding of smells in restaurant/shops, dispensers of room smell refreshers upon detecting unpleasant aromas (gym locker rooms, bathrooms 😊)

· In Space: my personal curiosity, How the Moon or Mars will smell if we could take the helmet off? Add EllieSmelly to the next space rover!

· Robotics: a humanoid (Wall-E) should have all or our senses!

· Education: combine EllieSmelly with application like Seek to record and identification of flora and fauna

· Historical: record the smell of events. Everyone has memories that are connected to a particular smell.

Design

EllieSmelly (upper left, Figure 1) consist of Adafruit SGP30, connected to PORT A of Edu-Kit. It has a custom made capture box to saturate and concentrate the smell within the chamber. The purpose of the capture box is to allow the sensor to have a fighting chance to differentiate aromas. It senses a smell and sends perimeter data to a Springboot application deployed on AWS Elastic Beanstalk using RESTful Services (see Figure 2). Note that this service could also be a Lambda application. The Springboot application stores and retrieves data from DynamoDB collections (see Figure 3 and 4). A curated collection should be used for retrieval (validated data of identified smells). ML can be implemented to create a curated collection.

The curated collection can be used to power the entire set of applications as described in the Story section.

Screens of EllieSmelly Edu-Kit Application

Credits and Disclaimer

The code for this prototype is heavily based on the provided Factory example. The back-end service is limited to what can be deployed in the free-tier Elastic Beanstalk. For example, I removed the desired dependency on <spring-data-dynamodb> to make my executable jar small enough to be deployed. Likewise, the free tier DynamoDB has a read/write capacity mode provisioned to just 5 records which was limiting.

I have problems with the HTTP client for FreeRTOS so I submitted samples using an external http poster. The SGP30 sensor is connected to and receives power from EduKit PORT A. (see Figure 5, above), however the readings are iffy at best.

In short, this prototype is still in progress but with school starting again for the kids I am out of time. I am submitting this prototype idea with the hope of Amazon and their unlimited resources can take it from here. I will continue cleaning up the code and getting these impediments working past submission date as my time allows.

I hope one day to have an Amazon EllieSmelly gadget in my kitchen to tell me if it is OK to cook the chicken that I purchased four days earlier. My kids and husband will appreciate that I don’t serve them cereal for dinner, even if I offer to heat the milk, to make it a “warm meal” that they were hoping for 😊

Looking Ahead

It is obvious that World 2.0 is emerging from this hard reset created by the pandemic. In the USA we are in the planning stages to improve infrastructure and manufacturers are busy producing a generation of smart vehicles for all forms of transportation. Most of the low hanging fruit products to support the yet to be realized smart infrastructure have not been developed. A smart mailbox, which will support a landing platform for delivery drones is a great place to start. I humbly submit this idea for your next contest. EduKit motor and sensing capabilities is suitable for development of a generation of smart mailboxes. The challenges presented by different home dwellings and delivery infrastructure makes this idea ideal for asking the world for their interpretation of how a smart mailbox will support their geographic needs.