So the essential question is: 

> How should we think about, or if possible compute, the semisimple
> support of a cuspidal character sheaf?

For example, let $G=SL_2$. We have the cuspidal character sheaf $IC(\mathcal{O},\mathcal{F})[1]$ where $\mathcal{O}$ is the regular unipotent orbit and $\mathcal{F}$ is the non-trivial $G$-equivariant local system on it. It has semisimple support $\left(\begin{matrix}1&0\\0&-1\end{matrix}\right)\in G^{\vee}=PGL_2$. In other words $IC(\mathcal{O},\mathcal{F})[1]\in\hat{G}_{\mathcal{L}}$ where $\mathcal{L}$ is the unique order 2 local system on the maximal torus of $G$. Do we have any method/algorithm to compute this semisimple support in general without a case-by-case study?

The most elementary way to realize the $SL_2$ case is probably that the Deligne-Lusztig induction of $\theta_{\mathcal{L}}$ (which is the character on a (any will do) maximal torus given by function-sheaf correspondence from a $\mathbb{F}_q$-structure on $\mathcal{L}$) contains two irreducible representations. The sheaf $IC(\mathcal{O},\mathcal{F})[1]$ then appears in the difference of their characters, again via function-sheaf correspondence. But this seems *ad hoc* as this decomposition relies implicitly on $G$ having disconnected center.

My apology for my ignorance if the answer is somewhere in the obvious literatures (e.g. [L1]-[L5]) that I fail to fully read. And thanks a lot for any answer and comment.