Motivation: There have been the instanton (anti-self dual connection) solutions to the Yang-Mills equation $d_A^\ast F_A=0$ which extremize the YM energy $\int_M|F_A|^2$, leading to the Donaldson invariants and even a Floer homology. There have been the monopole (connection + spinor) solutions to the Seiberg-Witten equations $D_A\psi=0$ and $F_A^+=\psi\otimes\psi^\ast-\frac{1}{2}|\psi|^2$ which extremize the Chern-Simons-Dirac functional, leading to the SW invariants and a nice Floer homology. These utilize the fundamental particles in the Standard-Model of physics… but not of General Relativity, where the gravitons arise.
So I would be interested in a Floer homology and/or invariants arising from gravitational instantons (Riemannian metrics), i.e. solutions to the Einstein Field Equations $R_{\mu\nu}-\frac{1}{2}g_{\mu\nu}R-g_{\mu\nu}\Lambda=0$ in vacuum (no stress-energy term $T$). Here $\Lambda$ is the "cosmological constant", which we may or may not want to assume is zero. Surely these have been studied extensively. (This term 'gravitational instanton' is used first (I think) in Stephen Hawking's seminal 1977 paper "Gravitational Instantons", and basic examples are the Schwarzschild and Taub-NUT metrics.)
Should I expect something to arise? Are there immediate obstacles? Otherwise this would have been done by now, right?
Downfall?: Perhaps the moduli space is too big, or boring, or unknown.
Progress?: Witten has even shown that (2+1)-dimensional gravity (with no cosmological constant) on $M=\Sigma\times \mathbb{R}$ (compact surface $\Sigma$) is an ISO(2,1) Chern-Simons theory, i.e. the equations of motion of the CS-action are precisely the field equations. And if there is a cosmological constant $\Lambda\ne 0$, then the same result holds when we replace the gauge group ISO(2,1) by SO(3,1) or SO(2,2) depending on the sign of $\Lambda$.
More: There is something to be said from Witten's recent paper "Analytic Continuation Of Chern-Simons Theory", but I am not ready to understand it.
Best Answer
Let me clarify a couple of issues from the previous answers/comments:
1) The linearization of $Rc-\tfrac{1}{2}Rg$ has mixed signs, and for this reason the equation $\partial_t g=-(Rc-\tfrac{1}{2}Rg)$ is bad, a kind of coupled backwards/forwards heat equation, i.e. no short time existence.
2) The linearization of $Rc$ is elliptic however (after fixing the gauge), and the Ricci flow $\partial_tg=-2Rc$ indeed is a good equation. In terms of functionals, it's better to consider the Perelman functional (whose gradient is $Rc$ up to gauge) instead of the Einstein-Hilbert functional (whose gradient is $Rc-\tfrac{1}{2}Rg$).
3) There was the issue of 1st order vs. 2nd order: While the YM-equation ($D^*F=0$) is second order in $A$, there is also the 1st order equation $F^+=0$ (antiselfdual connections). Solutions of this 1st order equation are special solutions of the YM-equation (as follows immediately from the Bianchi identity). There is a similar story for Ricci-flat metrics: Here the "special" solutions are the ones with special holonomy. Having holonomy $SU,Sp,G_2$ or $Spin_7$ implies that the metric is Ricci-flat. E.g. by solving the first order system $d\psi=0,d*\psi=0$ for a 3-form $\psi$ on a 7-manifold you get Ricci-flat metrics with holonomy $G_2$. Indeed, there is a proposal by Simon Donaldson for a higher order gauge theory based on $G_2$ (though, instead of just counting $G_2$-structures, the idea there is actually to "count" the number of associative submanifolds...)