For the "Megatrends" special area in Hall 16 at Interschutz 2020, we had a bold idea in February 2019: How can we simulate and make the undetected exponential spread of a virus at enormous speed in a small area accessible to our visitors?
With anonymous motion data on the surface, parameterizable pathogens, a selected patient 0 and a final visualization. After a few rough PoCs (Proof of Concepts) we got a sense of feasibility and set the scope. And began to develop. We had no idea that our experiment would become real very soon.
The first challenge is to capture a truly anonymous motion profile. Current approaches (e.g. in the Corona app) rely on installed apps that exchange data for contact tracking via a Bluetooth API newly developed by Google and Apple. This makes perfect sense in terms of user numbers and the area to be monitored. However, our application is much more manageable in terms of area - and was a little too early to benefit from the technical development.
In order to record the movement data on the special area anonymously, we first divided our area into quadrants.
Figure 1: Floor plan with grid approximation
A Raspberry Pi is hidden in (almost) each quadrant, and each of these microcomputers performs a BLE scan at regular intervals. The BLE Ids and signal strength seized with it are pre-sorted, anonymized and aggregated locally: on the one hand, we receive a baseline for signal quality across the surface from our own stationary devices, and on the other hand, visitors who leave our test area or only briefly strip can be sorted out again..
Figure 2: Capturing the movement data
At regular intervals, the data is further aggregated and smoothed and transmitted to our message bus in the backend. The amount of data transferred is significantly reduced before transmission, so that a commercially available PC on the display area can be used for further processing and display. At the beginning, two main points spoke against a cloud deployment: the generally poor Internet connection at trade fairs and the sheer number of messages sent. In the meantime, we were able to optimize at least the last point away, the message volume was massively reduced and comes especially with a cast number of collected data points. Only the quality of the uplink speaks against a theoretical use of the cloud, but we continue to focus on the "on-premise" use.
Our backend consists of a message bus, a ASP.NET core component (Blazor) and a database for perusing the spread and storing the time series. As soon as a simulation is running, the time-dependent data points are spatially located on the surface. Unfortunately, the signal strength of the data points is strongly dependent on the current operation on the surface, so that calibration is necessary. This is where our own Raspberry Pies come to the rescue, as we know their distance and Tx Strength and can use it to calculate an approximate signal strength baseline. The location then takes place in the meter range and is therefore more than sufficient for our purposes. It is worth noting that for the further protection of the privacy of visitors, we do not work with real-time data, but for our simulation only time series that are older than 10 minutes are used.
Figure 3: Components of the architecture
At the start of the simulation, a participant becomes Patient Zero. We have planned to provide the participants with a specially registered BLE beacons - since it is no longer possible to assign a specific BLE Id on the special area to a participant by anonymization.
Figure 4: Start of the simulation with Patient Zero
In addition, certain parameters of the pandemic pathogen are adjusted: transmission distance, incubation time, transmission time and time to recovery - whereby in our case we generously assume that a recovery also corresponds to immunization.
Figure 5: Configure the virus
Figure 6: Patient Zero runs off and starts the infection
The participant now moves freely through the exhibition space, takes part in other experiments, conducts discussions with other visitors or exhibitors and then returns. Recording stops - and our pandemic explorer starts working. The current time series of patient Zero as well as historical time series of visitors are mixed and used as a basis for the simulation of the spread. Together with the participant (Patient Zero) we discuss visualization of the spread before the entire simulation is reset. However, the original idea of gamification of the spread was subsequently dropped - in view of the tragedy of the COVID-19 pandemic.
Figure 7: The infection spreads
Figure 8: The infection is spread through the Super Spreaders
Figure 9: Spatial isolation has protected few participants from infection
We will make this experiment available to visitors as an demo via FUTURA in the coming weeks and hopefully also on Interschutz 2021 in Live.