I’d measure that before I would believe it since I doubt you have a half-ohm-plus resistive drop across all of that wirng, unless something is REALLY strange and/or wiring has been added that is too thin.
If that’s the case, the diode drop is the least of your worries.
These contacts and leads should be able to carry the full load at full rated voltage/current with losses that are truly de minimus - and a couple of tenths of a volt, (or so), is not de minimus.
Back when the Pi cam out, the voltage drop with many of the USB A to USB A micro wires I bought, used with the “5v 2A” adapters I bought was on the order of 0.1-0.2v.
The last adapter I bought says “5.3v at 2.5A” to make the voltage at the end of the wire meet the USB spec.
The wires in the cables I have are like soldering to hair.
I forgot to mention to run it from an remote ssh shell so that you can still see the output when the Pi gets the power jerked away.
Today’s Run:
tillShutdownLIfe.py with LIDAR not spinning, GoPiGo3 just standing, RPi3B+ load 0.04
- Battery protection circuit shut off at 5:57:30 8.28v Battery 7.47v GoPiGo3 reading
- “Too Low” 9.12v Battery 8.31v GoPiGo3 reading 5:55:40 (~2 min reserve)
- “Nearing Shutdown” 9.72v battery 8.91v GoPiGo3 reading 5:43 (~14 min reserve)
I’d still take a good, hard, look at that wiring.
A 0.3v drop is unacceptable.
Assuming a full-load edge case, that’s over one watt of power loss in the cable. The several tenths of an ohm in series with the varying load of a 'bot can cause voltage fluctuations I’d rather not see.
Before you do anything else that requires power, I would strongly suggest you beef up the wiring, get top-end cables that can handle the current draw, and get that voltage drop as deep into fractional milliamp territory as possible.
It is not unreasonable to expect a full-load voltage drop near the absolute minimum you can measure.
I do the same when installing automotive batteries. I get everything sparkling clean and shiny, clamp down tight, and measure the voltage drop while cranking:
Anything more than fractional millivolts gets investigated and corrected.
That doesn’t sound like a “One SMOP problem” (Small Matter Of Programming).
Convenience wins this time.
You might want to put a fractional-ohm dummy load on the system to pull max power.
If you can get the end-to-end drop into the fractional milliamp range at full rated power, (check the drop on both sides of the line), then maybe add a snubbing network from V+ to V- to improve S/N ratio, and things should be peachy.
Don’t forget to periodically re-test the static drop at full load.
Power source points typically have mega-gain from VCC/VSS to the signal leads where line noise is concerned.
Microvolts of noise across VSS/VDD can turn into significant fractions of a volt of noise in signal circuits causing all kinds of weird things to happen.
You might want to do a voltage drop and/or noise study on your own 'bot running balls to the wall at max current draw.
I didn’t understand a word of that beyond “You might want to”, and it is quite ok because the power system seems robust enough to proceed with software configuration, unit testing, and learning ROS.
What I did do is send this email to service@talentcell.com:
I’m considering keeping 10.0v as my “Warning - Battery Low” but raising the safety shutdown to 9.75v providing roughly 20-25 minutes of warning before the safety shutdown monitor pulls the plug on my playtime (which will announce “System Shutdown Scheduled In 2 Minutes” ).
With that level, I can safely issue a “sudo shutdown -c” and know that I can push everything up to git and then re-issue the shutdown.
And they responded within 3.5 hours - wow!
**BUT as I discovered there is only 2 minutes between 9.15v and sudden death, so I’m going to see how many cycles I get with a 9.75v shutdown. Hopefully 700 cycles of 5 hours initially ending with 2.5 hours should last me more than a year, and would end up being $0.008/hour ($35/4375h).
Carl’s batteries cost out around $0.003/hour, but hey, it’ll be a ROSbot.
A very good emperical study.
What I am trying to say is that as power requirements rise, wiring and connector losses become more significant, and as current draw increases, this turns into possibly significantly reduced playtime and undesirable connection heating.
In my entire power-electronics lifetime, I have personally seen only one battery fire/explosion1 compared to dozens if not hundreds of fires/damage instances caused by poor contact and underrated wiring.
There were certain desirable results:
My brother’s dog now knows better than to jump on my lap if I have a soldering iron out.
My brother now respects that “I need to concentrate” means that bothering me can cause undesirable things to happen.
Aside from scaring the poop out of both of them, nothing really serious happened.
I am sure I will revisit the “total playtime measurement” and “15 minute warning orange LED” once ROSbot gains the ability to start the LIDAR spinning and is wandering around in the resultant map. (Estimate 3-6 months to learn ROS and teach ROSbot to use it.)
Good to know “no dogs were hurt in this portrayal of real life”
What that is trying to say is:
Assume an audio amplifier driving speakers at several watts.
A tiny amount of power-supply noise, (microvolts, even fractions of a microvolt), will show up as huge amounts of amplified noise at the speakers.
This is why high-end amplifiers spend a considerable amount of their design cost budget on the power supply.
At least a third of the design effort expended in any reasonable PCB circuit design is power supply filtering, routing, and noise isolation.
Even the Raspberry Pi has dozens and dozens and dozens of special filtering capacitors and inductors scattered all over the PCB just to control this.
Summary:
It’s better to use thicker wire and very solid and tight connections.
If you want to open a separate thread, I will be happy to discuss and suggest good power transmission designs and effective filtering methods.
There is a reason HP took away my screwdrivers when they switched my job description from mechanical engineer to software engineer forty-five years ago. My robots run on “Velcro”.
Amazing - I ended up exactly where I guessed six days ago.
That’s actually quite good. It verifies that your original assumptions were dead-on.
Getting to the same conclusion from multiple logical starting points shows that the result is trustworthy.
The specific starting point was your discharge curve that showed the mid-point of the knee to be around 3.25v per cell. Thank You.
(after a little clean up)
I remember seeing a picture of Pablo Picasso’s hands (in Life magazine?) many years ago and they were all wrinkled and mottled.
A picture of my hands looks more like the way the bottom of your feet feel after square-dancing on Lego bricks.
In my case, I’ve never been afraid to try things. Though one of the things in the top right-hand drawer of my workspace desk is ointment and Band-Aids. 
I also keep a supply of 3-M “Command” velcro fasteners.
Tip:
Stores like Home Depot sell rolls of double-sided velcro plant-tie ribbon, (hooks on one side, loops on the other), that’s dirt cheap.
If you don’t mind the green color and the relatively narrow width, it’s excellent for strapping things together. (I’ve been using it to organize network cables for decades.)
You get several feet of it for a few dollars, as opposed to the “official” Velcro cable ties at Micro Center for dollars each.
Also! Avery “return address labels” are great when wrapped around a cable as a way to tag it.
I also sent the following - CC’d to MR support.
Let’s see what they say.
Thanks. This is an area where I’m not nearly as concerned about the details as either of you. I’ll mess with it just enough to get by with what I need. I’m certainly leaving performance on the table by not exploring the boundaries of the hardware envelope, but there’s only so much time in the day. Besides - I can just ready your and @cyclicalobsessive 's posts to learn about that stuff 
/K
Let’s close this thread - Marking it [SOLVED]