Troublesome Quartic Angle

GROMACS version:2020.7
GROMACS modification: No

I am trying to use a quartic angle to model the backbone of a polymer which has two minima at 100o and 150o with a barrier of ~ 8 kJ/mol occurring at the mid-point between these wells, at 125o (Note: I am using the MARTINI forcefield, so the angle is defined between three beads).

It is my aim for this angle to switch between these two states over the course of the simulation. To determine the coefficients of the polynomial I solve a matrix of five linear algebraic expressions using a python script and I am confident that my input coefficients are correct (I plot these to ensure they have both real roots and conform to my expectations of shape – see attached image).
I am encountering some strange behaviour:

(1) If I start my simulation in the first energy well (e.g., 100o) I find that in minimising the system this tends towards 160o-180o, which initially implied that the barrier was too high.

(2) however larger barriers and higher temperatures still yield (100, 300, 500, 800 and 1000 K) the same outcome. e.g., angle tending towards and stabilizing at ~180o with virtually no sampling of the other well.

I am puzzled by this outcome and am keen to get ideas/suggestions/guidance of things to try to get this to work. There appears to be limited discussion of the quartic angle in the user group, and it seem to be a useful feature of Gromacs which might be of use to others in the community.