Electronic – “Motor PWM” vs “LED PWM” on the ESP32

Shoot-through current refers to the condition where both switches/MOSFETs on one side of the H-Bridge are on simultaneously. Under normal conditions, the H-bridge is under one of the following conditions:

If, however, both switches on one side are on simultaneously, a huge current can flow (Only 34mΩ per leg, remember?), which is usually destructive. I'm not sure why both switches would be on - Possibly an issue with gate capacitance and switching time?

The datasheet says that the PWM pin is:

Voltage controlled input pin with hysteresis, CMOS compatible. Gates of low side FETs are modulated by the PWM signal during their ON phase allowing speed control of the motor.

You might get better luck with higher voltages if you modulated the PWM such that it was centered on the ON-pulses:

Motor on:     ----    ----     --      --
S-Left-Top__/------\________/------\________  
S-Rght-Bot___/----\___________/--\__________
S-Rght-Top__________/------\________/------\
S-Left-Bot___________/----\___________/--\__  

But you'll definitely want to look very carefully at Figures 4, 5, and 6 of that datasheet.

Your frequency is way too high, or conversely, you are not waiting long enough between pulses. You should be sending a 1 to 2 ms pulse every 20 to 50 ms. Note that this means 50 Hz maximum.

Also, your "1.7 V offset" makes no sense. These pulses should generally be 0 to 5 V digital signals. I followed the link you provided, but that only goes to a product sell page not a datasheet, so we can't tell exactly what voltage levels the digital signal must be. When in doubt with hobby servo units, use 5 V logic.

Try a 1 ms pulse every 20 ms. That should send the arm to one end. Then try 2 ms every 20 ms. That should send the arm to the other end. Try it with the 3.3 V signal straight out of your processor and see if that works. If it doesn't, make it a 0-5 V signal instead. That really should work if everything else is hooked up right. If you don't have a level translator chip, you can rig up something with a NPN transistor and pullup resistor. That will invert, but just invert the processor output to compensate. A HCT (note the T) logic gate powered at 5 V can also be used as a 3.3 V to 5 V converter.

Are you sure the processor power is properly filtered so that the current spikes the motor draws doesn't effect it? A schematic and scope traces with a clear description of what happens in each case would help a lot.