Unfortunately the short answer is probably not easily.
Define “physically approach a node”, are we talking 1m, 1cm, 1mm?
If you get within the near field of the antenna (a few wavelengths) then you will impact the antenna gain pattern, if you get close to the line of sight path then you could attenuate the signal or create reflections which destructively interfere with the signal and so cause reflections to be picked up instead of the LOS signal.
So depending on what exactly is happening noise spikes are to be expected. Plus UWB radio is very low power and is trying to measure signal times to fractions of a nanosecond, even under ideal conditions it’s going to be a little noisy. You add in all of the RF noise in a typical room and you’re going to get the occasional bad measurement.
The simple wins all involve getting the best possible signal conditions. Avoid putting the units hard up against walls, keeping electronics / housings a little away from the antennas, keeping things off the floor, and ensure you have the best antenna orientation. You’ve already done all of that.
There may be a very slight gain by raising things up from the desk surface a little more and possibly rotating 180 degrees about the vertical axis so the component sides are facing each other. But I’d expect any improvement from that to be minimal at best.
Using PANS there isn’t a lot more you can do to improve the performance other than improve the physical setup which is already close enough to perfect. All the other settings are locked down in order to maintain the approvals certification.
You can improve overall system performance by adding some processing to the measurements. Any large errors will always result in distances that are too long (there is one exception to this but it’s fairly rare). So if you reject any measurements that represent a large jump in range and then apply a low pass filter or averaging to what’s left you can get a better range number at the cost of a slightly slower response to changes in range.
PANS is generic, as an off the shelf system it’s great but since it’s generic it’s never going to be optimised for any given situation. If you only want point to point range then different firmware will let you measure the range much more rapidly giving more points to average / filter. While this doesn’t remove the noise directly it means you can filter while maintaining responsiveness and do it all on the hardware so that the output from the device is cleaner. However since this involves custom firmware it’s significantly more effort and development work, especially if you need to certify the final result.
Which brings us back to the short answer, while you could in theory improve on this there probably isn’t much you can easily do to improve things in any meaningful way.