I assure you: commercial players are working very hard on this. The difficult part is actually power management. There are clearly very sophisticated algorithm

I assure you: commercial players are working very hard on this. The difficult part is actually power management. There are clearly very sophisticated algorithms that can do no meal announce closed loop management of t1d (just google scholar "closed loop type 1" -- it's a very popular problem for control systems researchers). But they take a lot of power. Embedded convex solvers for large MPC schemes do not come cheap, especially when you want them running every 5 minutes! I have used this DIY loop system in the past. It is extremely power hungry and requires recharging daily, even when plugged in half the day. And I don't even think they are doing anything exotic like MPC. I stopped using it because of those battery issues and the implementation is gnarly -- it's basically a collection of bash scripts and relies on the operating system (armbian linux) to schedule doses. No RTOS, no watchdogs... Power management is extremely important in a pump. If a pump dies every <24 hours without being recharged, that impacts both the patient experience and can be very dangerous. If it dies overnight you get no insulin and you will be very sick the next morning. That said, I love that this is a thing. It pushes the tech forward and gets people excited about a machine cure, which is the only viable solution to "curing" type 1 at the moment.