It seems like it shouldn't be a big deal to design a constant power supply for a rover. It just needs the same standard current sensor that every RC plane has. 3D Robotics actually donated a current sensor with integrated 5V regulator. The problem is the rover is too small to fit it anywhere.

Sadly, regulating power by making PWM proportional to voltage didn't work. The idea was higher voltage at lower duty cycle would be equal to lower voltage at higher duty cycle, but power still increased with increasing voltage. The motors aren't constant resistance for all voltage & there's an unknown power usage from automated steering. It wasn't as effective as a linear regulator, which was still not true power regulation.

3DR uses an INA169 current sensor, which is an op-amp calibrated to convert the voltage difference across a 0.0005R resistor into a known voltage range. Their current range is 90A, while the rover needs 2A.

That leaves building something out of a 1R resistor with an op-amp. The maximum voltage drop should be 2V with a maximum current at 2A, but it's actually PWM with the maximum current an unknown higher amount for a shorter duty cycle. Then the PWM result needs to be smoothed.

They limit the bandwidth with a .1uF & 110k described in the datasheet. Using 1/(2piRC) gives 14Hz. Your best bet is to probe the resistor output & smooth it as much as possible with a big cap.

Constant power remanes a bridge too far for something which also needs a suspension system. Some waveforms showed nothing abnormal.

Deriving PWM purely from voltage, no load, at 12.5V we have .19A 2.38W

At 10V, we have .20A 2W

At 9V, we have .24A 2.16W

As usual, the frequency response of the probe makes suspect voltage ranges.

With constant PWM for all voltage, we have

12.5V .43A 5.3W

10V .37A 3.7W

9V .33A 2.97W

So voltage derived PWM is better than nothing, in the normal voltage range.