I found it has to do with the accuracy of floating point numbers (like the IEEE 754 doubles).
Jussi once taught me:” They look like real numbers but they are only a finite subset of them. Near zero this subset is so dense that one can perform simple arithmetic operations with decent accuracy. But the accuracy of operations gets smaller as operands get further from zero – because the set of doubles gets so sparse.”
The model “to big”.
I scaled the whole by factor 0.001, and everything goes fine. drone animation 001.3dm (7.1 MB)
Moving the objects closer to the World’s origin (0,0,0) would even improve the matter, I guess.