I learned about this before from the Herzog film Encounters at the End of the World (recommended) where they show how in a blizzard it is very difficult to navigate since without visual cues humans will walk in circles.
This doesn’t appear as a big mystery to me. It’s very difficult to move in a straight line; one has to maintain a very delicate balance, either between the amount of force exerted by the left vs right foot or the position of the steering wheel in the case of driving the car. Humans (and from what I can tell any system based off of a neural network) have a very hard time with “absolutes” but rather think and act in relative terms. Neurons fire due to changes in stimulus; a sustained stimulus results in neuron firing rates that die down over time. My hypothesis is that people make one tiny mis-calibration in pressure while trying to move in a straight line and after a few seconds any sense that such a mis-calibration occurred has been ignored by the brain since the change in force itself was so small and short-lived. The human then goes on unknowingly at the new level of force (slightly longer right food stride, steering wheel tiled 2 degrees to the left, etc) resulting in slow and gentle curvature over time.
A major piece of evidence for this seems to be that while the studies constantly show that humans will travel in circles, there’s no consistency regarding how tightly or quickly the circles appear. This would be explained by the fact that in such an experiment, each person (and even in each separate trial by that person) the deviation from “absolutely perfect” force exertion occurs at a pseudo random time and for a pseudo random amount.