Can you clarify what you mean by "in1 has nothing"? In the case where there are zero records on the input the Modify Fields node runs successfully:
Similarly, if some records have non-Null values the conversion is successful:
The one situation that I have found that generates an error is where all of the values in the field are Null:
If this is the case you mean then you could consider using a Split node to separate records where the field value is Null. The records with Null values can then be input into a Modify Fields node where the format attribute for the unicode to double conversion is set to f
The records from the Split node with non-Null values are input to another Modify Fields node where the format attribute is not set - allowing them to be converted to double type, as normal.
The two sets of records can then be combined using a Cat node.