The forward kinematics of rtb.models.Panda() is wrong. This can be easily verified by comparing the Denavit-Hartenberg values or just looking at the zero configuration:
panda = rtb.models.Panda()
print(panda.fkine(np.zeros(7))
Not sure if you solved it or not. The difference is because the rtb assumes the panda hand is attached, so an extra 0.104 mm is subtracted from 0.926, and the hand has 45 degrees of offset.
The forward kinematics of rtb.models.Panda() is wrong. This can be easily verified by comparing the Denavit-Hartenberg values or just looking at the zero configuration:
which yiels
and not