You can see the source data of the visualization you're showing as an example (flights over Queens) here, as the dataset is public.
What they are using is a ton of points for each different position through which you see "the flights" moving, and a
timecode column, through which they are able to say Torque: "This point will appear in this moment of time".
In your case, you can follow something similar. Imagine you have 3 different locations that you want to track (each location has different coordinates). For each location, you would have to have a table like:
the_geom, name, your_order_column, your_size_column
(0,0), Location1, 1, small
(0,0), Location1, 2, medium
(0,0), Location1, 3, big
(10,10), Location2, 1, small
(10,10), Location2, 2, medium
(10,10), Location2, 3, big
(10,10), Location3, 1, small
(20,20), Location3, 2, medium
(20,20), Location3, 3, big
You can set here a Torque category map, which gets as time column "your_order_column" and as category column "your_size_column". Then, your 3 locations will appear always and will change sizes with respect of time.
I have created the following example for you. The map is available here and the dataset is here.
Also, the CartoCSS code I'm using is as follows:
/** torque_cat visualization */