I generally put in antiwindup to my written PIDs. I helps get you back to reality more quickly.

Keith Cress
kcress - http://www.flaminsystems.com

It doesn't take a book. Most sources are wrong anyway. The so call clamping technique is something that gets passed by word of mouth without and critical thinking or understanding.
First calculate the control term from the P and D terms. Call this Upd.
Now when calculating the integrator term, Ui, limit like this
Ui(n)=max(min(Ui(n-1)+Ki*error(n)*Δt,100-Upd,-100)-Upd)
Now the total control output,U(n) will never exceed 100%
U(n)=Ui(n)+Upd
Notice that if Upd is big then Ui(n) may be pushed back into negative values.
That is OK if your system is tuned right.
See page 6 of 27 where this is applied. I didn't plot the integrator term but you can see
on the next page that implementing the integrator limit in this way allows the control output to saturate but still not cause over shoot. This is an old Mathcad worksheet where I was trying to impress the importance of the derivative term.
I was not trying to show how to limit the integrator or I would have plotted that too.
http://deltamotion.com/peter/Mathcad/Mathcad%20-%2...
This is gold

Peter Nachtwey
Delta Computer Systems
http://www.deltamotion.com
http://forum.deltamotion.com/

Generally to avoid windup, I detect when the PID has lost effective control of the system (loss of sensor, loss of actuator etc.) and then place the PID in manual of freeze it. Perhaps in your case detect when the robot is lifted up and "freeze" the controller?

Hello

google Adaptive Cruise Control System by January Mahlalela, I have explained and implemented the back calculation method.

Regards