How I Make My First P.O.C

Everything is going really fast since the start of this project. But after only two months of hard work, I start the development of (kind of) the Proof Of Concept (P.O.C) !

I will not explain here what is a Proof Of Concept or what is it for, but I will try to explain how I proceed in my very specific case.

1. State of the art

Since I consider a DCOP algorithm as part of my solution for a long time, I first reviewed a lot of different algorithm, trying to find the best one for my project. I take a look on ADOPT, DPOP, even some specific ones like CoCoA. Finally, I choose the DPOP algorithm (Dynamic Parameters Optimization Problem) which gives me more advantages :

• It is one of the fastest in the execution time. It is always a nice advantage.
  
  
• All agents are ranked with a DFS Tree which allow agents to organize themselves during the process. 
  
  
• Also, the DPOP algorithm is a 100% decentralized method since all agents executes the same code : there are no “intelligent mediator” to manage them. This is really convenient since the system does not rely on a central process.

Coupled with this research phase, I made my own inquiries about the medical field, and more precisely, nurses work. My objective was to gain enough information about their process to define a clear purpose for my project. And to do so, I needed to understand their needs.

2. Mathematical aspect

With this first step, I define the main goal of my system :

“Avoid syringe pump to ring without involving nurses too often”

Now that it was said, it was time to start the tough part : I needed to translate my constraints in a mathematical language in order to transcribe it inside the DPOP algorithm. This is required because the algorithm optimize constraints through a (mathematical) minimizing function. Therefore, I need to describe my constraints as functions. Above are some examples.  

For Mi = {m1, …, ml} the set of devices linked to the agent i, if the agent has no devices, then there is no need to call the nurse.

This next function transcribe the following constraint : If two agents are in the same geographical area (= they are neighbors), they can eventually synchronize themselves in order to avoid two interventions in a t_synchro laps of time.

To understand those functions, notes that vi is the hypothetical time that remains before the next nurse passage in the room. For instance, if vi = 5, it means that the next passage of the nurse is encouraged in the next 5 minutes. Thus, the algorithm try to find the best vi affectations for each agent i of the system depending on those constraints (by trying to minimize their results).  

3. Hands on keyboard

With those constraints, I finally start to code ! Here is a photo of the current installation.

I am working with two raspberry pi : each one of them is an agent – which mean that my DPOP algorithm is running on both of them in the same time.

I also have an AVNET Linux server which run a “server” specific code. This process is here to give a kind of Let’s go! signal to all agents. This Let’s go! allow all agents to start the DPOP algorithm in the same time. This is just for implementation. Maybe I will remove it later, when my agents will be more advanced.

For those who are interested, I code this algorithm in Python 3, and my agents/server are communicating with an MQTT Protocol.