A local community for delivering meals to COVID-19 patients sick at home

I want to share what we are doing here with a group of like-minded people, in the hope that others can take advantage of our approach and learn from our mistakes.

We are starting a system by which people who are sick with COVID-19 at home can order very cheap meals (2 meals a day for $10) and get deliveries every day from volunteers.

We made deals with restaurants to agree to serve “rations” (i.e. dinner + lunch next day) for $10/day. The consumers, who have already registered on our Discourse instance, must call them in, order rations for the next day, and pay for the order. Every day, our drivers will go to the restaurants and pick the orders and deliver them.

Here, in the Bay Area, it is almost impossible to get two meals for $10, particularly delivered. So it is a real deal.

I cannot give a link yet because we have not officially launched. But we are hosting on Discourse, and all the workflow and communications occur on Discourse.

We are developing some custom code to pull out customer info, do automatic driver routing (traveling salesman) etc. We will put all of our code on Github as it gets completed.

We will also share our protocols and procedures, along with lessons learned, publicly, so that others can do what we are doing.

20 Likes

Thanks for letting us know, a case study and ideas to make Discourse work better for this use case are totally welcome.

Feel free to reach out to people on Calling out for volunteer Community Managers 📣 for help.

15 Likes

That is awesome to see the power of the human :heart: and spirit! Keep up the great work!

6 Likes

Glad to see others are looking at this, I’ve been made aware of another group that has been asked to handle 26k meals a day and they’ve just approached me for solutions. :astonished:

Looking forward to seeing the outcomes, it may be that this is immediately applicable within that other project and could save me a bunch of headache.

9 Likes

26K meals per day delivered?

The delivery logistics for this volume are tough. We are assuming that our volunteer drivers can deal with 15 rations, i.e. 30 meals per delivery trip, and we are giving them each an optimized driving route. It still takes them about 1.5 to 2 hours to complete their run, a duration that we feel is sustainable 2-3 days per week for our volunteers.

For us, 26K meals would translate into 13K rations. We would expect that this would mean about 7K stops (about 2 rations per stop) for us. Assuming volunteer drivers and regular cars for delivery, that would mean, for us, 700 drivers driving every day, so, for us, in reality about 2,000 drivers (because ours won’t volunteer to drive every day).

If you have control over the source of the meals (we don’t make the food so it is much harder for us), then you just need to partition the address list between drivers, then optimize each route. Our volume is a lot lower than yours (we are thinking 2-3K meals per day, 1/10 of yours), so we don’t auto-partition the address list. Each delivery address is assigned to a neighborhood, and our dispatchers hand partition the delivery list by neighborhood (it is a simple spreadsheet sort). Then we optimize each driver route through a web server running on AWS, and give each driver an optimized driving route (simple traveling salesman algo).

We have considered auto-partitioning the full address list. But we dont need it yet, and we don’t have much money so we want to keep our AWS and Google Maps API tabs low :slight_smile: For the generalized multiple vehicle routing problem, there are simple algorithmic solutions that we could use, for instance here:

https://developers.google.com/optimization/routing/vrp

However, with that many deliveries, I imagine you must cover a very large geographic area, so possibly you also have to look at your drivers’ home addresses, which takes you out of the regular VRP. Possibly simple heuristics can partition the drivers’ groups and the delivery addresses into subproblems where both the deliveries and the drivers are in the same geo area, in which case you can just use a canned VRP solution again.

Btw we use the GMap travel duration as the point to point distance.

5 Likes

Not large thankfully, think ultra dense. A large number of unionized workers became unemployed very quickly less than two weeks ago. They ran food banks for days, but it’s even less practical because it funnels thousands of people through a small space and any asymptomatic carriers have the potential for massive fallout. The majority of drops will be to homes, but there will also be several last-resort centers which will cater for staggered groups.

8 Likes