From the rules:
A robot will be considered to be controlling a token if it is touching a robot or contained within the robot. The Judge’s decision is final.
In other words, if the token is “within” the robot, as determined by The Judge, it does not need to be touching the robot directly. Additionally, there is likely going to be leeway for cases where the cube is obviously under the robots control (such as if you somehow manage to stack cubes on top of your robot).
For reference, I have rendered a simulation of the event using cutting edge modelling technology, to illustrate what does and does not count.