A **real tree** is a metric space $(M,d)$ satisfying the following two conditions:

(1) for every $x,y\in M$, there is an unique isometry $\phi$ from the closed interval $[0,d(x,y)]$ onto $M$ such that $\phi(0)=x$ and $\phi(d(x,y))=y$; and

(2) any one-to-one continuous mapping $f:[0,1]\rightarrow M$ has the same range as the isometry $\phi$ associated to the points $x=\phi(0)$ and $y=\phi(1)$.

Metric spaces that can be isometrically embedded into a real tree are exactly the ones who satisfy the **four point condition**, that is, metric spaces $(N,\rho)$ such that
$$\rho (a,b)+\rho (c,d)\leq max (\rho (a,c)+\rho (b,d),\rho (a,d)+\rho (b,c)) ,a,b,c,d\in N.$$

My question is: what is the most natural/useful way of defining an n-dimensional version of a real tree? Is there any hope to obtain a nice characterization of subspaces, in the spirit of the four point condition?

I would be please to see an answer to this question even for two dimensions.