Planetary Gearbox Trouble and Optimization
 

Hello everyone, I would appreciate some help/advice on how to go forward with this mechanism.
I recently designed and 3d printed two variations of a planetary gearbox for a heavy load (Potentially around 120nm of torque). The final design won't be 3d printed, but metal based.
The first variation is:
0.03 backlash for all gears (planet, sun, internal, etc)
The second variation is:
0.03 backlash for only the internal and sun gear.

When I assembled a very crude prototype, with only the gears lying face down with no other external structures holding them in place, the first variation had a lot of wiggle room, and was more "noisy" and vibrated more than the second variation.

I'm not very experienced in how that might translate to a "real-time scenario" where the mechanism would be used for a huge load. The use case is in a robotic arm, for the base and shoulder joint.

Is the first variation the more optimized choice, or would the second variation be better?
Would adding external structures help, such as adding shafts to hold the gears in place, adding a planet carrier, etc?
Any improvements I could make?