I agree with Christiaan. Similar movements are stored nearby and are thus difficult to distinguish (see How are movements stored in the brain? ), even more so with the spatial resolution of EEG. What I would like to add is that for imagery to be motor imagery, you have to imagine the movement in first person perspective. Imaging someone else to execute a movement will not show in the motor cortex. This is a nice paper showing result of novices using a new BCI.
BCIs go further than steering a Robot or a pointer on the screen (which can be achieved in a similar fashion). BCIs are also used for disabled peopled who cannot speak for instance. For this they have developed a P300 speller: a matrix of letters which one by one flicker at a high speed. Whenever the letter that you were thinking of highlights in the screen, a strong P300 can be recognized in an ERP. In this paper, multiple algorithms are compared for its accuracy.
A final way of BCI makes use of the frequency spectrum (see this paper for example). By focussing more or less, relaxing more or less etc, the power in the different frequency bands will vary. At our university we have a game called mind-ball, a very simple one-vs-one game where you have relax as much as possible (have high alpha and theta) to move a little ball to your opponents side of the table.