I’m assuming you’re operating your circuit as a voltage divider, fed into the ADC. I’d bet money you found that the digital value you are getting is operating within a range, and not between 0 and 255 (8bit) or 0 and 1023 (10bit). Of course you can identify your operating range and scale it to fit your output (PWM) which from what I read is 0 - 255. That will linearly sync your measured resistance from fully relaxed to fully flexed to your ESC output.
The downside to using this method is that since you are effectively “zooming” in on a small section of your ADCs measurable range (0v to 5v in the case of your Nano there RIP my two, I fried them two days ago and was forced to change platforms), your output will become very jumpy due to insufficient resolution on your input.
I haven’t looked into the ADC on the Nano very much to see if it even has this, but the proper way to handle a voltage divider based input is to have an ADC with +VRef and -VRef inputs.
On an integrated board such as the Nano, this would mean using up an additional two analog inputs to be used as your VRef inputs, however you would be able to take two trimmer pots (one for the +VRef and the other for the -VRef) and then the ADC channel actually reading the flex sensor, will only measure the voltage range you want.
There are ways to do this completely by potentiometer but its a shitload of tweaking, it’s not very reliable, and there are other drawbacks such as excess power consumption. The E-INK displays I’m using on my design have to be driven with RAM restrictions on the 328 AVRs. I just got two ESP8366’s in today that have only one analog input and that means I can’t directly use one on the glove side of the LCS. Since I’m using SPI as my main communication bus for the radio and display, I figured why not use a 4ch 12bit SPI chip…
Hope this helps you tackle some of the problems I think you may encounter… I have about 20 years of experience building electronic systems so let me know if you need any guidance, I’d be more than happy to help. Two people tackling the same problem is still a 100% increase in efficiency overall.