I have a Garmin Fenix 6 Pro and I find it pretty accurate. I set the watch to watch for sleep between 10pm and 8am. Recently with COVID-19 I've been going to bed past 2-4pm. In the morning the first thing I do is check the sleep stats. I find them very accurate. They correctly track my pit stop runs in the middle of the night as well. Today I woke up at 11.43am and the watch got it right. Interestingly after I wake up and lay in bed for a few more minutes, while I'm still in bet, I also start to get bazillion of notification from phone on my watch. I feel very confident about the watch's ability to track the start and the end of sleep right. I cannot tell about the REM time but it seems within the correct ball park.
There were some rough nights where it showed no REM at all. The truth is that I was pretty low energy in those next days and not at my full potential. However, when the watch tells me I had >2h of REM, I'm an energy bomb.
The body battery is another thing that is quite magical in indicating my resources. Two days ago I found myself at 10/100 midday. At 4pm my concentration and deep thinking ability was almost inexistent.
From the engineering perspective I noticed that the SP02 sensors attached to the finger, to the smart watches and other devices tracking pulse, O2 and heart functions they use pretty much the same technology to read the pulse, blood oxygenation levels etc.
There is a lot of software and signal engineering that lays on top of the sensor readings that differs on each manufacturer. To get these settle to accurately predict REM and VO2 and other more advanced activity labelings they do need a lot of data and iterations. Fitbit and the Apple watches went through thousands of data points and they do seem to be miles away now.
So far I found that with each software update the Garmin Fenix is getting a lot better at these predictions. Certainly for all these devices there is a learning curve and we are quite frankly at the beginning of it.