Back in the pool today with the new polyfuses. Despite bumping these up to 10A each, I was still seeing the thrusters reset sometimes - so what was going on. Well looks like a few things. First, when I disable then enabled the balance system (that's the PIDs which keep the craft positioned in the water according to the IMU and user's requirements) I wasn't clearing all the old data out of the PIDs. This would sometimes mean a sudden change thrust from 0 to something quite big. This appears to generate a current spike which resets the thruster controller. Second, if I physically pushed the craft violently around an axis, the PIDs attempts to correct this would also result in a large current spike.
The software fix for this was to make sure I reset *everything* when I restarted the balance system, and also limit the rate at which a thruster could change speed, so limiting the potential for current spikes. Both these fixes worked very well and the craft now balances correctly.
But ... I can't help but feel I really have a hardware issue here. Power electronics is not my domain, but I'm wondering if sudden changes in the amps being consumed by the ESCs is causing problems for the nearby PWM chip. While that's on a separate power supply and has basic decoupling, I wonder if the shared ground is as issue in some way?