Consider a continuous surjection $f:[0,1]\to[0,1]\times[0,1]$.
It can be proved that set of self-intersection points must be dense.
In the Hilbert curve, the set of self-intersections are points (a,b) such that either a or b can be written as $\frac{m}{2^k}$ for some integers $k≥1$ and $1≤m<2^k$ (see this explanation).
So the set of self-intersections is dense and of measure $0$, but you cannot draw any vertical or horizontal line without intersecting it.
This leads me to the question: can there exist, for some space-filling curve, an arc (homeomorphic to a non-degenerate closed interval) $\tau\subset [0,1]\times[0,1]$ such that $\forall_{t_1,t_2\in [0,1]} f(t_1) = f(t_2) \in \tau \implies t_1 = t_2$? That is, such that $\tau$ includes only non-self-intersection points?
No, there can be no such arc.
Suppose for the sake of contradiction that $g$ is an arc in $[0,1]^2$ that contains no self-intersections of $f$.
Let $g \colon [0,1] \to A$ be a homeomorphism.
Then $A$ is compact and connected.
Since $A$ is compact and $[0,1]^2$ is a Hausdorff space, $A$ is closed.
Since $f$ is continuous, $B = f^{-1}[A]$ is also closed, and therefore compact.
Now the restriction of $f$ to $B$, $f \restriction B$, is a continuous bijection.
A continuous bijection from a compact space to a Hausdorff space is a homeomorphism, so $B$ is homeomorphic to $A$.
Thus $B$ is also connected.
A closed, connected set of real numbers is a closed interval, so $B$ is a closed interval $[p, q]$.
Let $D = ([0,1] \setminus B) \cup \{ \min B, \max B \} = [0,p] \cup [q, 1]$.
Then $D$ is obviously compact, but $f(D)$ is not:
$f(D)$ is the square with the arc $A$ removed and then just two points put back, so any other element of $A$ is a limit point of $f(D)$.