How much headroom do I need on the 5V rail for 8 servos?

If you're worried about voltage sag during servo stalls, the placement of your bulk capacitors is probably more critical than the exact number of ceramics you put on the board. Since servos pull massive current spikes, you want the capacitor closest to the servo headers—meaning the path impedance should be as low as possible. The large electrolytic capacitor you placed near the servo headers is the right idea for providing the necessery energy buffer during a sudden high-draw event, but make sure the trace leading from that cap to the servo rail is extremely wide and short. It needs to handle the transient current without resistive loss. Regarding the rail sag, even with good decoupling, if the total spike current exceeds the steady-state capacity of your buck converter, you'll still see drops. Since the servos are pulling power, consider a dedicated high-current, low-inductance trace or even a small local buffer cap right at the servo connector. This isolates the servos from the primary power delivery noise and ensures they have enough juice when the gait gets unpredictable. On the charging while running setup, the use of a power mux like the LTC4416 is definatly the right approach to isolate the battery pack from the VBUS line. When the system is powered by USB, the MUX correctly switches the power source to VBUS, allowing the charger to operate uninterrupted while the main system runs off that line. The pack sitting unconnected is exactly how these chips are designed to operate in this mode; the mux handles the switching logic and isolation for you. Your layout strategy around the power paths seems solid, but focusing on the trace width and minimizing loop inductance around the bulk caps will address the instability you're concerned about. Don't underestimate how much impedance matters when dealing with transient loads like those produced by stalled MG90S servos. It makes a huge difference in system stability.

I'm running a similar setup but I am more worried about the noise coming off the switching regulator itself. Even if the power rails stay stable, I keep wondering how much interference that heat and switching create. shold I be adding any filtering components or maybe shielding around the section where the power converter sits, since it’s close to the main controller? I haven't looked into that part of the design yet and feel a bit lost on where to start.

Totally agree about focusing on the trace impedance near the servo headers. That low loop inductance thing is critical. I learned this when I was building a diferent actuator system, and the difference between a large and small capacitor footprint made a massive difference in stability during peak current draws. I had to physically shrink the copper path to see the voltage ripple acutally drop. It’s all about keeping that current path super short.

I agree that focusing on the physical traces is key. But when you are dealing with high current spikes from servos, the ground plane underneath those power paths is just as important as the wire itself. I once designed a controller for a heavy-duty lift where I specified a full internal ground pour that was 4oz copper just to handle transient surges. By keeping that return path solid, we kept the voltage noise under 5mV even when the motors stalled. Make sure your ground connections to the switching regulator are as wide and direct as possable to keep the current from spreading out across the board.

I've run into this same issue where the power dip from the servos doesn't just hurt the servos, but it makes the main controller choke. Even if the rail stays above the minimum, the noise spikes bleed into the sensitive 3.3V regulator feeding the microcontroller, causing random resets. I had to physically seperate the servo power path and the brain's power path completly on the board. Keeping the high current flows away from the sensitive chips makes all the difference in stable operation.